2.1. Transition to split funding modes and the rise of competitive project funding

The transition from exclusive block funding of research organisations to a split funding mode that combines block funding with project grants is one of the most important changes in the governance of research. Although the initiation and speed of this institutionalisation varied greatly between countries, its impact on research in many fields has been severe. In fields that depend on external project funding, researchers compete not only for recognition of their research but also for the opportunity to continue research at all. In the governance system, research funding agencies have emerged as powerful new actors that affect the direction of research.

These changes have enjoyed considerable attention in science studies. In particular, funding agencies have been studied in some depth because of their unique role and growing influence in the science system. They were considered as intermediary organisations that mediate between science policy, scientific communities, and researchers who apply for and receive grants [Braun Reference Braun1993]. Later, the concept of “boundary organizations” was introduced by Guston [Reference Guston2001] in order to emphasize their active role in mediating between policy and science (see also Kearnes and Wienroth Reference Kearnes and Wienroth2011). Principal-agent theory has been suggested as a theory that can explain structures of and processes in and around research councils [Braun Reference Braun1993; Rip Reference Rip1994; Guston Reference Guston1996; Braun Reference Braun1998; Van der Meulen Reference Van der Meulen1998; Braun and Guston Reference Braun and Guston2003; Caswill Reference Caswill2003], but has been criticised for not being able to do justice to the complex embeddedness of research councils [Morris Reference Morris2003; Shove Reference Shove2003].

The analyses of research councils as intermediary organisations include only very few general comments on their influence on the content of research. Braun [Reference Braun1998] makes a convincing theoretical argument that it is possible to influence the cognitive development of science through research council funding (see also Rip Reference Rip1994). How this influence occurs, and how specific changes of research content are achieved, has not yet been explored in any depth. Instead, research has focused on the process of selecting proposals for funding. Changes in research content have been treated indirectly as outcomes of these selection processes, if at all.

The consequences of decision-making by peer review for the content of research have been a concern of science studies ever since the first study of peer review at the nsf, which found the likelihood of having a grant approved did not differ from a random selection (Cole et al. Reference Cole, Cole and Simon1981; for a recent review see Van Arensbergen et al. Reference Van Arensbergen, van der Weijden and van den Besselaar2014). Later discussions held peer review decisions to select “excellent mediocrity”, i.e. good but not excellent research, mainstream research, and low-risk proposals [Chubin and Hackett Reference Chubin and Hackett1990; Travis and Collins Reference Travis and Collins1991: 336; Horrobin Reference Horrobin1996; Berezin Reference Berezin1998].

The increasing relative scarcity of research funding has triggered an “excellence turn” in science policy, namely attempts to fund research with the potential to transform or significantly advance its field. These attempts manifest themselves in new funding programmes, whose implementation challenges peer review to identify exceptional research (which is variously termed “excellent”, “breakthrough”, or “performative”). Contrary to the above-mentioned findings and beliefs that peer review is incapable of promoting exceptional research, appropriately “conditioned” peer reviews appear to be able to identify exceptional research. Studies of decision-making procedures in peer review have identified the conditions and practices supporting the identification of exceptional research [Dirk Reference Dirk1999; Guetzkow and Lamont Reference Guetzkow and Lamont2004; Heinze Reference Heinze2008; Luukkonen Reference Luukkonen2012]. Funding programmes that utilise peer review for the selection of “excellent” research proposals have been shown to be able to identify research leading to scientific innovation or research that was categorised as “excellent” by independent experts [Lewison Reference Lewison1999; Lal et al. Reference Lal, Hughes, Shipp, Lee, Richards and Zhu2011; Wagner and Alexander Reference Wagner and Alexander2013; Laudel and Gläser Reference Laudel and Gläser2014].

Researchers were also interested in the responses of peer review processes to the thematic heterogeneity of grant proposals. Within reviewers’ fields of expertise, Travis and Collins [Reference Travis and Collins1991] observed “cognitive cronyism”, i.e. a bias of assessors towards their own scientific perspectives. Lamont [Reference Lamont2009] and Huutoniemi [Reference Huutoniemi2012] identified several strategies that members of interdisciplinary panels use in order to achieve consensus. In interdisciplinary social sciences and humanities panels “deferring to expertise”, i.e. the delegation of decisions to panellists who belong to the fields addressed by a proposal, appears to be common [Lamont et al. Reference Lamont, Mallard and Guetzkow2006]. Langfeldt [Reference Langfeldt2001] observed discipline-specific interpretations of general guidelines for evaluations.

A separate, more recent strand of bibliometric research turned to the question whether peer review does indeed select the best applicants [Bornmann et al. Reference Bornmann, Wallon and Ledin2008; Van den Besselaar and Leydesdorff Reference Van den Besselaar and Leydesdorff2009; Campbell et al. Reference Campbell, Picard-Aitken, Côté, Caruso, Valentim, Edmonds, Williams, Macaluso, Robitaille, Bastien, Laframboise, Lebeau, Mirabel, Larivière and Archambault2010]. Unfortunately, the methodology of these studies seems questionable for three reasons. First, they contrast peer review decisions on grants with bibliometric measurements (publications and citations) of applicants’ pre-grant or post-grant performance, which provides very little information unless one is prepared to accept bibliometric indicators as more valid performance measures. Second, successful applicants are often compared to unsuccessful applicants, i.e. a control group that is inherently problematic [Neufeld and Hornbostel Reference Neufeld and Hornbostel2012]. Third, reducing the independent variable in studies of success to the award of one particular grant does not do justice to the complexity of funding situations. It is not entirely clear what this kind of study contributes to our knowledge about the impact of funding councils on research.

In contrast to this emphasis on selection processes, only very few studies actually investigated changes in the content of research due to the (anticipation of) peer review decisions. The only major exception is the link between competitive grant funding and research performance which has been investigated primarily in the context of evaluations of funding programmes. Studies of grant-funded individuals and research centres provide mixed evidence [Gaughan and Bozeman Reference Gaughan and Bozeman2002; Jacob and Lefgren Reference Jacob and Lefgren2011; Neufeld and von Ins Reference Neufeld and Hornbostel2011; Neufeld and Hornbostel Reference Neufeld and Hornbostel2012; Neufeld et al. Reference Neufeld, Huber and Wegner2013; Bloch et al. Reference Bloch, Graversen and Pedersen2014; Langfeldt et al. Reference Langfeldt, Benner, Sivertsen, Kristiansen, Aksnes, Brorstad Borlaug, Foss Hansen, Kallerud and Pelkonen2015]. It turns out that researchers receiving fellowships and “centers of excellence” do not necessarily produce better research than comparable researchers or centers without such funding. The only interesting finding in this context is that the award of prestigious grants is not associated with increased research performance but is associated with increased career success [Böhmer and Ins Reference Böhmer and von Ins2009; Bloch et al. Reference Bloch, Graversen and Pedersen2014].

An important observation that challenges the whole idea of competition for funding leading to better research was contributed by Butler and Biglia [Reference Butler and Biglia2001] in a study that, unfortunately, is difficult to find. Butler and Biglia found that while Australian research funded by grants from the National Health & Medical Research Council did indeed have a higher citation impact than unfunded research, research that did not have to rely on grants at all because it was funded by block grants had the highest impact [ibid.: 13-14]. In a similar vein, Auranen and Nieminen [2010: 831] find no clear causal link between the degree of competition in a science system and its publication performance and efficiency.

The few studies that attempt to identify thematic changes caused by research council funding are based on interviews and mostly rely on behavioural changes reported by interviewees. These studies appear to confirm an earlier observation that the “fundability” of a proposal is firmly integrated in considerations of “do-ability” [Fujimura Reference Fujimura1987]. Researchers respond to priorities set by funding councils and to their perception of peer review being biased towards mainstream and low-risk research [Morris Reference Morris2000; Gläser et al. Reference Gläser, Lange, Laudel, Schimank, Whitley, Gläser and Engwall2010; Leišytė et al. Reference Leišytė, Enders, Boer, Whitley, Gläser and Engwall2010]. Although window dressing plays an important role in attempts to secure funding, the thematic priorities and decision-making practices of funding councils do have an impact. Both the selection processes and their anticipation by researchers result in the latter orienting their research more towards mainstream, low risk, and applied topics. In addition, grant funding appears to hinder changes of research topics because selection processes favour proposals that are linked to applicants’ prior expertise. Researchers must “bootleg” money for the start of new research under the cover of existing grants [Hackett Reference Hackett1987: 143; Morris Reference Morris2003: 364-365; Laudel Reference Laudel2006: 496; Gläser et al. Reference Gläser, Laudel, Lettkemann, Merz and Sormani2016].

Thus, while grant funding increases the flexibility of a science system and supports the concentration of resources on the best performers, little is known about its effects on the content of research. Adherence by researchers to topics that can be expected to be funded may reduce the diversity of fields on the macro-level, may force them into specific directions, and may hinder rapid innovation. The “excellence turn” in science policy can be interpreted as a response to this concern. However, too little is yet known about the “pathways to impact”, the actual adaptation of research by researchers and research groups, and aggregate effects of such adaptation. For example, we know very little about those who lose in competitions for grant funding. What happens if researchers who need grants to conduct their research are unable to win any? There are some hints at low-level research of reduced validity and reliability [Laudel Reference Laudel2006; Gläser et al. Reference Gläser, Lange, Laudel, Schimank, Whitley, Gläser and Engwall2010] but, by and large, we just do not know.

2.2. Incorporation of public policy goals in science policies and the rise of targeted funding

Since the middle of the previous century, science policy has begun to incorporate public policy goals in an attempt to increase the contribution by science to solutions of societal problems [Remington Reference Remington1988; Behrens and Gray Reference Behrens and Gray2001: 179-180; Lepori et al. Reference Lepori, van den Besselaar, Dinges, Potí, Reale, Slipersæter, Thèves and van der Meulen2007; Hessels et al. Reference Hessels, Grin and Smits2011; Berman Reference Berman2012; Reference Berman2014].^{Footnote 1} Roughly at the same time, science outgrew the opportunities to fund its growth, and science policy began to face decisions about what science to fund [Cozzens Reference Cozzens1986]. As a consequence, expectations for science to provide specific knowledge content have been significantly enhanced, and have been inscribed in evaluation and funding procedures. Funding councils have incorporated such priorities into their procedures for allocating project finance in various ways (see e.g. Kearnes and Wienroth Reference Kearnes and Wienroth2011 on the responses of the UK’s Engineering and Physical Sciences Research Council to state expectations). In addition, many states have become more directly involved in supporting research in particular areas for policy purposes. This is why effects of this trend on the content of research are partly included in considerations of the impact of project-based funding (2.1), and higher education reforms (2.4) and academy-industry links (2.5). In this section, we consider the impact of state investments in “emerging fields”.

The trends described above created the need for science policy to fund research selectively. Scientists responded to this new funding strategy by casting their fields and topics as particularly promising. The intersection of these trends has been described as the emergence of “strategic research” and “promissory science”. The former concept refers to “basic research carried out with the expectation that it will produce a broad base of knowledge likely to form the background to the solution of recognised current or future practical problems” [Rip and Voß Reference Rip and Voß2013: 41; see also Van Lente and Rip Reference Van Lente Harro and Rip1998; Rip Reference Rip2002]. The latter concept was used by Hedgecoe [Reference Hedgecoe2003] to refer to the promises researchers make in order to attract support for what they rhetorically construct as a promising field. This echoes the observation of a “promise-requirement cycle” according to which promises made by scientists in order to secure funding become institutionalised as expectations on which fields have to deliver [Rip Reference Rip1997: 628-632]. Rip also pointed out that the attempts to raise state interest and employ science policy in the funding of research has furthered the use of “umbrella terms” (terms of unclear scope) because these terms are useful in gaining support for scientific enterprises that cannot yet be precisely described. Umbrella terms also serve to suggest integrated efforts where they do not exist. In her study of materials science, nanotechnology and synthetic biology, Bensaude-Vincent [2016: 54] confirms that researchers in these interdisciplinary “fields” “still remain strongly grounded in their referent disciplines” (for nanotechnology, see also Marcovich and Shinn Reference Marcovich and Shinn2014).

This basic constellation has been studied for many emerging “fields”, “technologies” or topics, among which nanotechnology is probably the most studied and synthetic biology is probably the most recent example [Molyneux-Hodgson and Meyer Reference Molyneux-Hodgson and Meyer2009; Bensaude-Vincent Reference Bensaude-Vincent, Merz and Sormani2016]. Science policy studies of emerging fields usually focus on the emergence of political priorities and funding instruments through interactions between scientists and political actors (e.g. Eisler Reference Eisler2013) or on the role of community-building funding instruments (e.g. Molyneux-Hodgson and Meyer Reference Molyneux-Hodgson and Meyer2009). The impact of the governance of emerging fields on the content of research has not yet enjoyed much attention.

While it is clear from studies of emerging fields that state interest contributes to the formation of fields and rapid growth, and that scientists attempt to exploit these new opportunities, we do not know how and with what results massive state investment changes the content of research. There can be little doubt that in addition to the window dressing triggered by targeted funding, such funding also increases research on the intended topics. Far less is known about the details of thematic changes. The few studies addressing responses to state funding hint at a complex picture that involves many intervening variables. For example, Leydesdorff and Gauthier [Reference Leydesdorff and Gauthier1996] compared responses by Dutch and Canadian scientists to “national innovation-oriented research programs”. They concluded that “Canadian researchers seem to have used the priority programs as an alternative source of funding, while their Dutch colleagues were able to use these programs to help their specialties grow above the national average, and in accordance with selected priorities” [ibid.: 448]. The authors ascribe these differences to the higher degree of integration of the Dutch research system and to “organizational slack (e.g. traditional or lump-sum financing)”, factors that made it possible for Dutch researchers to follow policy signals more closely with additional research, while their Canadian colleagues were forced to use the additional money for maintaining their existing lines of research [ibid.].

The inevitable indirect effects of political priorities in research funding are also still to be explored, in particular studies about knowledge that is not produced due to the foci created by politics (see also 3.3). Almost 20 years after Leydesdorff and Gauthier, a study by Laudel and Weyer [Reference Laudel, Weyer, Whitley and Gläser2014] showed that the high integration of the Dutch science system remains but the system now operates without slack. Under these conditions, state priorities can create a quasi-market failure that makes non-priority fields disappear. On the micro-level, Smith [Reference Smith2010] describes the perception by health inequality researchers that the pressure to produce “policy relevant research” limits their autonomy and creativity.

A particularly interesting situation emerges when the state simultaneously perceives the necessity to promote a field and to regulate it, as has been the case for research using human embryonic stem cells. Taken together, two studies by Furman et al. [Reference Furman, Murray and Stern2012] and Brunet and Dubois [Reference Brunet and Dubois2012] provide an opportunity to analyse the impact of state regulation on the content of a rapidly growing field. Furman et al. use bibliometrics to study the response of US science to the 2001 decision by the Bush administration to enable federal funding for research with existing human embryonic stem cell lines, to prohibit federal funding for the development of and research on new cell lines, and to place no restrictions on the use of other than federal funds for research with human embryonic stem cells. The authors observe a decline in research with human embryonic stem cells after 2001 until 2003. The subsequent recovery between 2004 and 2007 is due to research at elite universities (which had easy access to non-federal funding) and international collaboration. The results suggest that while the governance intervention led to a change in research content for researchers in some universities, these effects were soon compensated for at the national level by the opportunities for other researchers to circumvent the governance instrument.

These observations are confirmed by a study of French regulation, which was based on the bioethics law and thus applied to all research using human embryonic stem cells. Researchers had to apply for permission, which was granted if the research was “likely to enable significant therapeutic advances” and “cannot be pursued using alternative methods of comparable effectiveness in the present state of scientific knowledge” (Art. L.2151-5 of the law, quoted from Brunet and Dubois: 263). Brunet and Dubois compare French research with human embryonic stem cells to its counterpart in the UK, and ascribe both the reduced scale of and the fragmentation of French research to the regulation. They attribute the reduced scale to the necessity of applying for permission for research, which made researchers avoid the field. The fragmentation (many groups not collaborating with each other) was caused by the demand for proof of therapeutic advances, which oriented the research towards specific diseases and applied questions rather than fundamental research. In contrast to their colleagues in the US, French researchers could not circumvent this regulation.

Although attempts by the state to create and direct research capacities for the support of public policy goals have increased significantly, the impact of these attempts on the dynamics of knowledge production has received comparatively little attention. The state influences research largely by implementing public policy goals in existing governance instruments rather than by creating new instruments. The impact of these policies on the growth of research and its direction appears to be significant but, again, systematic knowledge regarding the link between specific governance practices that are applied under specific circumstances and have specific effects is missing.

2.3. Evaluations and the rise of performance-based funding

Since the late 1970s, many countries have altered their block funding of higher education institutions by replacing or supplementing input-based funding for research by performance-based funding. The logic behind these changes includes incentives (to reward better performance), redistribution of resources (to increase efficiency), and improved management (to provide necessary information for change). Performance is measured either by peer review (which is often informed by quantitative data) or by quantitative indicators. The information about research performance is then used as an input for funding formulae that allocate resources in a zero-sum game (see Whitley and Gläser Reference Whitley and Gläser2007; Auranen and Nieminen Reference Auranen and Nieminen2010; Hicks Reference Hicks2012 for overviews).^{Footnote 2}

The effects of performance-based research funding schemes have attracted a great deal of attention from both stakeholders and researchers. Political discussions and research were centred on the same questions: Do the new funding schemes contribute to improving research performance? And do they have unintended negative effects? These questions are of particular interest to our review because they have been addressed by quantitative and qualitative investigations in higher education research, the sociology of science and science policy studies.

Most studies focus on system-level effects. Studies of the Spanish, English, Australian, Norwegian and Danish performance-based funding systems asked whether the introduction of new funding schemes changed research performance, and with what effects. The analyses applied a quasi-experimental logic that interpreted the introduction of performance-based funding as treatment, and causally ascribed changes occurring after the treatment to these schemes. Research performance was operationalised in terms of publication behaviour (publications in “good” journals, however determined) and citation impact of publications.

The major challenge faced by this approach is the control of all possible confounding variables. This challenge is most clearly illustrated by a discussion about the Spanish funding scheme, which provides increases in salaries for those academics who publish a certain number of articles in international peer-reviewed journals. Jiménez-Contreras et al. [Reference Jiménez-Contreras, De Moya Anegón and López-Cózar2003] claimed that the introduction of this scheme in 1989 caused a disproportionate increase in the number of Spanish articles in such journals because the increase occurred after the change in the performance-based salary increase scheme was introduced. The authors analysed possible confounding variables––investment in R&D, numbers of researchers––and concluded that the limited growth in these variables could not explain the increase in publication performance. They dismiss several other factors such as international mobility, international collaboration, and participation in European funding programmes, as well as the cumulative effect of policies.

The critique of this analysis by Osuna et al. [Reference Osuna, Cruz-Castro and Sanz-Menendez2011] is based on a review of five threats to the internal validity of analyses of time series [ibid.: 580-581]. They found the factors dismissed by Jiménez-Contreras et al. to be possible causes if time lags between funding and publication were taken into account, and demonstrated that a control group of researchers who were not subject to the performance-based funding scheme showed a similar increase in research performance. Most importantly, their discussion illustrated the problem of confounding variables: there were too many such variables to consider for a causal ascription of the effects to be possible.

The situation is even more complicated when performance-based funding schemes for universities are studied. The complex causal chains linking these to national-level changes in research performance include the strategic responses of universities to the funding schemes, the resulting changes in the situation of individual researchers, the strategic responses by researchers to their changed situation, and the aggregation dynamics of individual changes in research and publication behaviour. Each of these steps is co-shaped by a host of influences on the behaviour of individual and collective actors.^{Footnote 3} Was the increase of low-impact publications in Australia caused by a funding formula that rewards number of publications [Butler Reference Butler2003], by the preceding higher education reforms that confronted a large number of academics with increased expectations to conduct research and publish (a change described by Meek Reference Meek1991), or by an increasing competition for project grants that emphasises track record, i.e. prior publication (a development described by Gläser and Laudel Reference Gläser, Laudel, Whitley and Gläser2007; Gläser et al. Reference Gläser, Lange, Laudel, Schimank, Whitley, Gläser and Engwall2010)? Is it possible to ascribe the increase in “good” Danish publications to the performance-based funding scheme [Ingwersen and Larsen Reference Ingwersen and Larsen2014]? The authors excluded changes in research funding, the increase in academic staff, and the internal dynamics of the Web of Science database as explanations because the growth in productivity and impact exceeds the other trends. They do not consider the combined effect of these trends or the increasing concentration of research funding, either as an effect of competitive grant funding or due to the new programme for the funding of “Centers of Excellence”, which was launched in the same year as the performance-based funding scheme [Langfeldt et al. Reference Langfeldt, Benner, Sivertsen, Kristiansen, Aksnes, Brorstad Borlaug, Foss Hansen, Kallerud and Pelkonen2015]. They describe but do not explain the observation that at least one university does not conform to the pattern.

Aagaard and colleagues conducted the most complex study of this type to date. They analysed the introduction of the Norwegian publication indicator [Schneider et al. Reference Schneider, Aagaard, Bloch and Noyons2014; Aagaard Reference Aagaard2015; Aagaard et al. Reference Aagaard, Bloch and Schneider2015]. The authors combined bibliometric studies, surveys, and interviews in order to assess the impact of this indicator’s introduction on research. “Research” was operationalised as publication behaviour. In addition, managers and academics at universities were asked to report changes in their organisation that they ascribe to the indicator.

As in the other studies, causal attribution appears to be difficult. The authors observe an 82% increase in “publication points” in the time between the performance measure’s introduction in 2004 and 2012 [Aagaard et al. Reference Aagaard, Bloch and Schneider2015: 109], and argue that this increase cannot be fully explained by other factors. In particular, they report that the number of researchers with a publication in the four main universities has increased by 116% while the number of R&D personnel has risen by only 5% [ibid.: 110]. At the same time, they report several observations that cast doubt on the causal ascription, including:

1. - the major increase in WoS publications (rather than “publication points”) occurring prior to the introduction of the indicator [Schneider et al. Reference Schneider, Aagaard, Bloch and Noyons2014: 547];

2. - substantial increases in R&D funding (37%) and R&D personnel (21%, Aagaard et al.: 110);

3. - “strong increases in publication activity among more recently established universities and among university colleges, many of which have previously had a much weaker focus on research” [ibid.]; and

4. - changes in the publication indicator itself [ibid.].

Aagaard [Reference Aagaard2015] reports an observation that supports the assumption of a causal impact on publication behaviour, namely the fact that the indicator is used in decisions about recruitment, promotion, and salary increases. Unfortunately, no timeline can be attributed to the introduction of such measures, which makes it impossible to link the “trickling down” of incentives [ibid.] to the changes in publication behaviour over time.

Not only are studies on country-level effects limited to changes in publication practices and citation impact (rather than research practices), they also appear to be liable to a “post hoc ergo propter hoc” fallacy, i.e. the conclusion that A must be the cause of B because B follows A.^{Footnote 4} However, attempts to meet the complexity of causation head-on by conducting qualitative studies face interesting problems of their own. These studies can investigate the actual translation of changing governance environments for academics into a changing production of knowledge. However, the observation of effects remains at the micro-level. In most cases, the causal chain under study ends with self-reported or self-announced behavioural change [Stöckelová Reference Stöckelová2012; Linkova Reference Linkova2014; Aagaard Reference Aagaard2015]. Lucas [Reference Lucas2006] did not ask about changes in the content of knowledge but focused on strategies of coping with the British Research Assessment Exercise at different levels in the university. Attempts to comparatively analyse researchers’ field-specific responses to performance-based funding schemes [Gläser and Laudel Reference Gläser, Laudel, Whitley and Gläser2007; Gläser et al. Reference Gläser, Lange, Laudel, Schimank, Whitley, Gläser and Engwall2010; Leišytė et al. Reference Leišytė, Enders, Boer, Whitley, Gläser and Engwall2010] found, not surprisingly, that for researchers in the sciences external funding was by far a stronger environmental factor than university funding and accompanying incentives. For example, Australian universities translated performance-based funding schemes applied to them in similar schemes applied internally. However, little money arrived at the researcher level simply because Australian universities were under-funded [Gläser et al. Reference Gläser, Lange, Laudel, Schimank, Whitley, Gläser and Engwall2010]. Researchers in less resource-intensive fields are less dependent on grant funding agencies and university management but may be susceptible to hierarchical pressure (see also Hammarfelt and de Rijcke Reference Hammarfelt and Rijcke2015).

Although they are able to explore researchers’ responses to their resource and governance environments, micro-level studies share some limitations. Interview-based studies can explore changes in interviewees’ problem choices as described in interviews. The exploration of research content in interviews is possible and can be triangulated with individual-level bibliometrics [Gläser and Laudel Reference Gläser, Laudel, Whitley and Gläser2007: 134-135; 2015a]. Nevertheless, two challenging questions remain. To what extent are reported changes in problem choices due to changes in governance? And do these changes add up to macro-level shifts in knowledge production? Answering the first question would require comparative ethnographies, whose resource demands exceed current funding patterns for social science research. The second question can be answered if detailed micro-level studies of changing research content can be linked to field-level changes by advanced bibliometric methods. Both types of methods have not yet been applied.

Taken together, studies on the impact of performance-based funding schemes on the content of research express a general dilemma of research on the impact of governance changes on research content. Studies on country-level effects, which are important in the research context of science policy studies, are forced to limit their measurement of research practices to indicators of publication behaviour and citation impact of publications, and must black-box the complex causal web that mediates the relationship between knowledge production, publication of new findings and their reception. Micro-level studies of changes in research practices and their causes can draw a more complex picture but are still largely unable to link their findings back to the aggregate dynamics at the levels of countries or scientific communities.

2.4. Higher education reforms and the rise of “new public management”

Over the past three decades, many oecd countries have introduced substantial reforms to the governance of universities [Paradeise et al. Reference Paradeise, Reale, Goastellec, Paradeise, Reale, Bleiklie and Ferlie2009]. So far, these changes have predominantly been studied by higher education researchers. They usually involved (and involve) the development of new relationships with the state and other extra-mural agencies as well as shifts in the internal management of academic institutions, which are often summarised under the generic title of “New Public Management” or npm [Schimank Reference Schimank2005; De Boer et al. Reference De Boer, Enders and Leišytė2007; Ferlie et al. Reference Ferlie, Musselin, Andresani, Paradeise, Reale, Bleiklie and Ferlie2009]. Against a common background of a growing discrepancy between performance expectation and public funding of academic research and teaching, states have increasingly sought to reconstruct universities as quasi-independent collective agents on which they can rely to realise national education and research goals.

These changes result in major reconstructions of authority relations in higher education. The increase in university autonomy from the state appears to be very limited at best in spite of the npm rhetoric that aims at shifting responsibilities from the state to universities [Westerheijden et al. Reference Westerheijden, De Boer, Enders, Paradeise, Reale, Bleiklie and Ferlie2009; Capano Reference Capano2011]. At the same time, the internal redistribution of authority from academics and their collegial decision fora to a managerial hierarchy of university senior management, deans, and heads of departments has progressed in many countries [Schimank Reference Schimank2005; Paradeise et al. Reference Paradeise, Reale, Goastellec, Paradeise, Reale, Bleiklie and Ferlie2009; Musselin Reference Musselin2013; Reference Musselin, Whitley and Gläser2014]. Evaluations of research performance have become one of the most important tools for state governance of universities and university management [Whitley and Gläser Reference Whitley and Gläser2007].

Although changes in university management significantly alter authority relations in universities, surprisingly little research has been conducted on the impact of these changes on research content. A theoretical argument by Whitley and Gläser [Reference Whitley, Gläser, Whitley and Gläser2014] points to principal limitations of universities’ abilities to “manage” research and thus to limitations of higher education reforms. However, there is little empirical research to support or refute this argument. Higher education research has described academics’ resistance to npm [Bauer and Henkel Reference Bauer, Henkel, Henkel and Little1998; Anderson Reference Anderson2008; Moscati Reference Moscati, Amaral, Bleiklie and Musselin2008] and the impact of npm on academic identities [Henkel Reference Henkel2000; Reference Henkel2005] but has done little to explore how npm affects the conduct and content of research. Studies investigating the impact of npm on research content at all focus on just one governance instrument––performance-based funding––and treat changed authority relations as a context of this relationship (see 2.3, and e.g. Morris Reference Morris2002).

Three exceptions should be mentioned. In their analysis of an evaluation exercise as a tool for profile-building at German universities, Meier and Schimank [Reference Meier, Schimank, Richard Whitley, Gläser and Engwall2010] ascribe the possibility of profile building to changed authority relations, to which the evaluations themselves contributed. The extent to which profile building changes research content could not be assessed at the time of the investigation. Laudel and Weyer [Reference Laudel, Weyer, Whitley and Gläser2014] link changed authority relations at universities and the tight integration of the Dutch science system to the possible disappearance of research specialties, to which profile-building activities of universities also contribute. Louvel’s [Reference Louvel, Whitley, Gläser and Engwall2010] analysis of shifting authority relations in French laboratories describes significant changes in the conduct of research but does not address research content.

The few studies of consequences of npm for research content again demonstrate the difficulty of causally ascribing changes in scientific knowledge to a particular governance process. The impact of npm on research content can only be assessed in the context of all other influences. So far, it seems to be rather weak in most countries because the opportunities by management to interfere with the recruitment of academics or to make them redundant are limited. In addition, opportunities for management to direct researchers’ choices of problems and approaches is limited due to the specific nature of scientific work [Musselin 2007; Whitley Reference Whitley, Engwall and Weaire2008; Whitley and Gläser Reference Whitley, Gläser, Whitley and Gläser2014]. This is why “governance by funding” is likely to remain a stronger influence than hierarchical steering.

2.5. Academy-industry relationships and the rise of privatised knowledge

At least four interrelated processes have contributed to changes in relationships between publicly funded research and industry. First, the state has significantly increased its demand for contributions by science to technological progress and economic growth, particularly through innovation [Behrens and Gray Reference Behrens and Gray2001: 179-180; Coriat and Orsi Reference Coriat and Orsi2002: 1493-1495; Kearnes and Wienroth Reference Kearnes and Wienroth2011; Berman Reference Berman2012; Reference Berman2014]. Second, the growth of scientific knowledge has increased the number of epistemic links between science and industry and thus of science-based industries [Mahdi and Pavitt Reference Mahdi and Pavitt1997; Koumpis and Pavitt Reference Koumpis and Pavitt1999]. This increases industry’s demand for public research support as well as opportunities for academic researchers to access relevant information, equipment and materials [Meyer-Krahmer and Schmoch Reference Meyer-Krahmer and Schmoch1998; D’Este and Perkmann Reference D’Este and Perkmann2011]. Third, the increased science base of many industries and its extension towards fundamental scientific research have increased opportunities for making profits by investing in research and either selling its results or betting on capital gains [Pavitt Reference Pavitt2001; Coriat and Orsi Reference Coriat and Orsi2002: 1500-1501]. Finally, the growing scarcity of public research funding has increased the willingness of researchers to seek funding from industry [Meyer-Krahmer and Schmoch Reference Meyer-Krahmer and Schmoch1998; Jankowski Reference Jankowski1999; Laudel Reference Laudel2006; Lam Reference Lam2010; D’Este and Perkmann Reference D’Este and Perkmann2011].

These trends made the volume and variety of links between university research and commercial interests grow and increased the latter’s influence on research content. We exclude from our consideration the major concern of science and innovation policy studies, namely the question whether increased academy-industry links contribute to innovation and economic success. Instead, we focus on consequences for the production of scientific knowledge by these researchers and their scientific communities that result from new incentives for researchers to engage in commercialization activities and the support by the state of forms that improve the opportunities for commercial actors to influence the content of publicly funded research.^{Footnote 5}

The impact of industry interests on the content of research can be considered at different levels and for different forms of academy-industry links. Levels include the content that is directly linked to industry interests––the content of collaborative or sponsored projects, more general research agendas of researchers having relationships with industry, and the knowledge production of scientific communities whose members engage in commercialization of research. At each of these levels, the different forms of academy-industry relations can exercise specific influence on research (Table 1). Investigations of the impact of university-industry links on the conduct and content of research included these forms in various combinations. A frequently used independent variable is “receiving support from industry” [Blumenthal et al. Reference Blumenthal, Campbell, Causino and Seashor1996; Behrens and Gray Reference Behrens and Gray2001; Evans Reference Evans2010a; Reference Evans2010b]. Murray and Stern [Reference Murray and Stern2007] and Campbell et al. [Reference Campbell, Weissman, Causino and Blumenthal2000]. investigated the impact of patenting research findings. Evans [Reference Evans2010a; Reference Evans2010b]. also included collaboration with industry as indicated by co-authorships.

Table 1 Findings on the impact of four forms of university-industry links on the contents of research on three levels of aggregation levels of research

Changes in the content of individual projects linked to industry have received little attention so far, probably for the reasons that explain the general avoidance of corporate research [Penders et al. Reference Penders, Verbakel and Nelis2009]. The only aspect of research content that has been explored is performance (see e.g. the review by Baldini Reference Baldini2008: 295-302). The studies reported by Baldini ask how research performance (measured by publications and citations) is associated with innovation performance (measured by patents or start-ups), albeit without establishing causality.

There is evidence concerning impacts of industry links on graduate student projects but, again, the evidence is mixed [Baldini Reference Baldini2008]. Some studies found no negative effects (e.g. Behrens and Gray Reference Behrens and Gray2001), while other studies reported by Baldini [2008: 302-304] point to enforced secrecy and reduced learning experiences. The only exception is research on the manipulation of outcomes in the sponsor’s favour. The suspicion that research projects funded by commercial enterprises are biased has initiated discussions and meta-studies in biomedical research. In his reviews of these studies, Krimsky [Reference Krimsky2013] concludes that “[i]ndustry-sponsored trials are more likely than trials sponsored by non-profit organisations, including government agencies, to yield results that are consistent with the sponsor’s commercial interest” [Krimsky Reference Krimsky2013: 582]. He also finds a small number of pharmaco-economic studies to show such an effect [ibid.]. Investigations of industry-sponsored research on specific commodities such as tobacco unambiguously demonstrate that this research is biased towards funders’ interests. Krimsky also points out that a variety of reasons may cause “funding effects”, and argues for ethnographies to determine the actual mechanisms at work between funding and bias. A study by Sismondo [Reference Sismondo2009] points to one of these mechanisms. Commercial services have emerged that design, schedule, and launch publications about drugs as part of pharmaceutical companies’ marketing strategies. Researchers from universities are asked to author or co-author such publications. Nothing is known yet about the impact of such “ghost-written” publications on a community’s knowledge production.

Beyond the immediate impact on the content of collaborative or sponsored research, the more general and potentially more sustainable impact of academy-industry links on academic researchers’ research agendas is of particular interest. Qualitative and quantitative studies have found sponsored research to influence research agendas by moving them towards more applied research or towards research with more potential for collaboration [Blumenthal et al. Reference Blumenthal, Campbell, Causino and Seashor1996; Lam Reference Lam2010]. The reverse also occurs: some researchers tend to “shy away” from problems associated with commercialising results [Owen-Smith and Powell Reference Owen-Smith and Powell2001]. Furthermore, the influence on research agendas works both ways: Cohen et al. [Reference Cohen, Nelson and Walsh2002] showed publicly funded research to have a large impact not only on project completion but also on the research agendas of industrial R&D across much of the manufacturing sector.

Academy-industry relationships do not only influence the agendas of researchers who have links to industry or commercialise their research findings themselves. Several studies have found that research agendas are influenced by the perception that too much knowledge of a field is controlled by intellectual property rights. Some researchers tend to avoid such fields [Eisenberg Reference Eisenberg, Cooper Dreyfuss, Leenheer Zimmermann and First2001: 225, 233; Walsh et al. Reference Walsh, Cohen and Cho2007; Murray et al. Reference Murray, Aghion, Dewatripont, Kolev and Stern2009; Evans Reference Evans2010a: 761]. The consequences of this intellectual migration away from intellectual property rights towards the directions in which fields evolve have yet to be investigated.

Research on academic spin-offs has found little change in university research agendas due to double affiliation of academics or collaboration with spin-offs. In their analysis of French spin-offs from cnrs laboratories, Shinn and Lamy [Reference Shinn and Lamy2006] distinguished three types of scientist-entrepreneurs but did not associate impact on university research agendas with any type. Zomer et al. [2010: 347] did not find Dutch research-based spin-off companies to alter the research agendas of university-based researchers directly but did not exclude the possibility of a “soft impact” through the academics’ changed awareness of practical problems.^{Footnote 7} In contrast, Cooper [Reference Cooper2009] found that the potential to create for-profit start-ups from research findings did influence problem choices. Unfortunately, he did not provide specific information about consequences for research content.

The impact of academy-industry relationships on the community level of knowledge production has been ascribed to aggregated changes in the conduct rather than content of research at the individual level. Concerning conduct, Blumenthal et al. [Reference Blumenthal, Campbell, Causino and Seashor1996] found industry support to be associated with increased secrecy and delayed publication. Campbell et al. [Reference Campbell, Weissman, Causino and Blumenthal2000] found that commercial activities were associated with data withholding.

The most complex study so far on relationships between industry collaboration, sharing and the diffusion of knowledge in scientific communities was conducted by Evans [Reference Evans2010a; Reference Evans2010b]. Combining qualitative and quantitative methods in an investigation of the Arabidopsis community, he observed that industry collaboration (both at organisational and individual levels) decreases the likelihood of materials sharing but increases the sharing of pre-publication manuscripts. Evans also found industry collaboration to slow or reduce the diffusion of knowledge but not to limit its thematic breadth (which he measured by calculating the annual proportion of citing articles that did not share coded scientific terms with the focal article). These results expand prior observations by Murray and Stern [Reference Murray and Stern2007], who found declining citation rates of research findings after patenting and thus a reduced diffusion of results. Evans's findings are indirectly confirmed by a study of research using genetically engineered mice. This use was first aggressively controlled by patent holders [Murray et al. Reference Murray, Aghion, Dewatripont, Kolev and Stern2009]. Murray et al. found a significant increase in the level and diversity of follow-on research after limitations on the use of these mice were reduced by nih agreements with patent holders. Further confirmation can be derived from observations of researchers’ “avoidance” of topics they associate with restrictions due to intellectual property rights [Eisenberg Reference Eisenberg, Cooper Dreyfuss, Leenheer Zimmermann and First2001: 225, 233; Owen-Smith and Powell Reference Owen-Smith and Powell2001; Walsh et al. Reference Walsh, Cohen and Cho2007; Evans Reference Evans2010a: 761].^{Footnote 8} Other scientists ignore or actively resist restrictions created by intellectual property rights [Murray Reference Murray2010].

Taken together, the studies investigating the impacts of academy-industry links on the content of research have established that such an impact exists, and that it varies between fields, but usually end there. This limitation is caused by two methodological problems. The first problem is the studies’ self-restriction to academy-industry links. Similar to the studies of performance-based funding schemes, studies of academy-industry links do not take into account other influences on project content or research agendas. They observe that, in most cases, collaboration with or funding from industry constitutes only a proportion of a researcher’s agenda but do not investigate how this proportion interacts with other influences. They also observe variations between fields but do not investigate the properties of fields that are responsible for the variation, let alone the ways in which these properties affect the building of research agendas in conjunction with academy-industry links. Since variations in the sharing of information about research are linked to the epistemic practices of fields.^{Footnote 9} These field-specific practices of information sharing constitute the background against which influences of academic-industry relationships on information and materials sharing must be assessed.

The second, connected, problem is the choice of empirical methods for investigating the impact on research content. Investigations are dominated by surveys and bibliometric studies, both of which are limited when it comes to exploring research content. The very few interview-based studies restrict themselves to asking about attitudes and behavioural change without exploring the consequences for research content. Ethnographies and in-depth explorations of research content in interviews are missing in the toolbox of studies investigating academy-industry links, which is why findings about research content are limited to the basic-applied dimension and to inconsistent findings about research quality.

The most promising results have been contributed by studies of knowledge diffusion. The creative bibliometric approaches by Evans [Reference Evans2010a; Reference Evans2010b] and Murray and colleagues [Murray and Stern Reference Murray and Stern2007; Murray et al. Reference Murray, Aghion, Dewatripont, Kolev and Stern2009] enable the observation of diffusion processes at the level of scientific communities. This level is crucial for many questions about research content but rarely addressed due to methodological problems. Extending the methodological ideas of diffusion studies to other problems of changing research content at the community level is certainly worthwhile.

2.6. Emerging trends in the governance of science

In addition to the main trends discussed in the previous sections we would like to highlight two emerging trends in the governance of science that have found the attention of science studies, namely the increasing activity of civil society actors in the governance of research and “governance by knowledge.” The dynamics of these trends is difficult to assess as yet. Both are linked to the emergence of new interests concerning contributions by science to solutions of societal problems or problems expected from science or science-based interventions. We limit our following discussion to those studies that include changes in research content, thereby excluding a large body of literature on public engagement with science and participatory science.

An increasing number of empirical studies has been devoted to attempts by new actors to influence the production of scientific knowledge. By “new actors” we mean collective actors that represent specific interests concerning science, which they believe to be insufficiently included in traditional governance arrangements. Examples include patient organisations, foundations concerned with public health or cures for diseases, and environmental movements.^{Footnote 10} We will consider them as social movements, although the interests are often represented by organisations that are not co-extensive with the movements themselves. These movements often identify knowledge that has not been produced (see also 3.3).

The new actors mainly use existing instruments in their attempts to shape research content. They lobby for a change of state priorities, mobilize resources for dedicated research organisations or contract research, or influence the agenda setting of research councils [Epstein Reference Epstein1996; Moore Reference Moore, Frickel and Moore2005: 305-308; Morello-Frosch et al. Reference Morello-Frosch, Zavestoski, Brown, Gasior Altman, McCormick, Mayer, Frickel and Moore2005: 263-265; Brown et al. Reference Brown, McCormick, Mayer, Zavestoski, Morello-Frosch, Gasior Altman and Senier2006; Frickel et al. Reference Frickel, Gibbon, Howard, Kempner, Ottinger and Hess2010]. Thus, their success––the actual change of research content they achieve––depends on the conditions already discussed in the previous sections, in particular the strong positions of researchers in the ultimate problem choices. For example, Pittens et al. [Reference Pittens, Elberse, Visse, Abma and Broerse2014] showed that although the involvement of patient groups in the setting of research agendas and the latter's translation into research programmes had a strong influence on these programmes, patient groups' priorities disappeared in the subsequent implementation of research programmes due to their exclusion from that phase.

Social movements also use less common channels such as directly influencing scientists, e.g. by building and maintaining networks of scientists working on topics they are interested in, or by befriending scientists [Panofsky Reference Panofsky2011]. Of these alternative means of influencing research content, we would like to briefly discuss influencing research content by providing knowledge. We consider this mode of influencing research content as particularly interesting because it utilises the way in which scientific communities intentionally and unintentionally shape the research of their members. Researchers select problems and approaches based on the state of the art, i.e. by taking in and evaluating the knowledge produced by their communities. Providing additional data may change the input to these selection processes and thus the content of knowledge produced by researchers. Although the contribution of data for research by non-scientists has a long tradition in a variety of fields (see e.g. Moore Reference Moore, Frickel and Moore2005), this rarely occurs with the aim of influencing the content of research. The shaping of research agendas through additional research information was observed in the case of patients and patient organisations [Brown et al. Reference Brown, McCormick, Mayer, Zavestoski, Morello-Frosch, Gasior Altman and Senier2006: 525; Panofsky Reference Panofsky2011: 38; Polich Reference Polich2011; Pols Reference Pols2013; Rabeharisoa et al. Reference Rabeharisoa, Moreira and Akrich2014]. While these studies confirm the effectiveness of this “governance by knowledge”, the actual change of research content brought about has not yet been sufficiently studied.

The same applies to similar attempts of industrial enterprises to “govern by knowledge.” Sismondo’s [Reference Sismondo2009] account of “ghost writing”––the preparation of research articles about the efficacy of drugs as part of marketing strategies––demonstrates that firms deliberately introduce knowledge in the scientific discourse. Although medical practitioners may well be their major target group, these articles are likely to have an impact on medical research as well. Unfortunately, this impact has not yet been studied. Firms also do the opposite by excluding knowledge from use. Again, the impact of this strategy on research agendas remains under-studied.

2.7. Strengths and weaknesses of effect-searching studies

The empirical studies discussed in the preceding five sections apply five distinct strategies to approach research content. A first strategy involves measuring macro-level effects such as speed or directions of knowledge diffusion. This strategy exploits citation links between publications, and links citation dynamics to other properties of publications or their authors. A second, frequently applied, strategy uses publication and citation indicators for measuring one particular aspect of research content, namely performance as expressed in volume or quality of research. This strategy, which attempts to draw conclusions about macro-level variables (the performance of the science system) from aggregate effects, seems particularly vulnerable in the light of the many problems associated with publication-based and citation-based indicators. Evaluative bibliometrics has been conducting an intensive discussion of these indicators’ validity and statistical reliability, field-specific behaviour, and necessary normalisation. Methodological advances that were necessary in the context of bibliometric evaluations do not always seem to be matched by the application of indicators in research projects on the governance of science.

A third strategy of measuring research content utilises publication indicators as proxies of publication practices. This approach has the advantage of matching the conclusions that are drawn to what is actually measured by the indicator. At the same time, it operates a step away from research content. Publication practices can be assumed to be linked to research practices and research content. However, the nature of this link is unknown, and is likely to vary across fields of research, career stage, and situations of researchers. Consequently, no conclusions about research content can be drawn from analyses of publication behaviour.

A similar argument must be applied to the fourth strategy, which attempts to measure research practices by surveys. Respondents are asked how their research practices (especially their choices of research problems) have changed.^{Footnote 11} This strategy suffers from a similar problem to that of exploring publication practices. Although practices of problem choice are more tightly coupled with research content than publication practices, the link between a general description of changing problem choice given by respondents to surveys, and changes in the content of knowledge produced by them, is far from clear. This is why statements like “research practices of these scholars were investigated using both publication statistics and responses to questionnaires” [Hammarfelt and de Rijcke Reference Hammarfelt and Rijcke2015: 64] simply do not ring true––whatever data were collected by these methods are unlikely to represent actual research practices.

Finally, a fifth strategy––using semi-structured interviews and ethnographies––enables conclusions to be drawn about research content because it can be used to explore with scientists their production of scientific knowledge in great detail.^{Footnote 12} This last strategy complements analyses of knowledge diffusion at the macro-level by exploring the reception and use of knowledge in the production of new knowledge and establishing the complex influences shaping knowledge production, which in turn can be traced back to governance. However, these studies leave us with micro-level effects which, albeit established with some validity, cannot yet be linked to the macro-level dynamics of knowledge. We still lack a “research content description language” in which (changes in) knowledge content can be described in a generalised way. Our vocabulary––“mainstream”, “low-risk”, “applied”, “interdisciplinary”––is too coarse to appropriately grasp effects like the disappearance or emergence of topics, methods, or kinds of data in particular research communities.

We are left with a dilemma, then. With few exceptions, studies of macro-level changes cannot convincingly establish causality due to the impossibility of measuring research content, the reduction of complexity enforced by quantitative methods, and the necessity of black-boxing processes of causation. Studies of micro-level responses, on the other hand, can address research content and establish causality in micro-situations without being able (yet) to link these changes to aggregation processes and macro-level dynamics.

3.1. Conditions for excellent research

The then famous internationally comparative study by Pelz and Andrews [Reference Pelz and Andrews1966] was probably the first to ask how organisations shape the conduct of research. Studies with a focus on organisational environments were conducted until the 1970s but were then marginalised by the constructivist turn. They were put back on the agenda by Law [Reference Law1994] and Vaughan [Reference Vaughan1999], neither of whom established a clear link between organisational phenomena and the content of research. In particular, although Vaughan convincingly demonstrates the fact that organisations “can complicate and manipulate the entire knowledge-production process” [Vaughan Reference Vaughan1999:931], her investigation of a single case of technological development cannot establish necessary or sufficient conditions for specific influences of an organisation on knowledge production, or explain how this impact would vary between technologies. The impact of organisations on the content of research is likely to vary with organisational structures and field-specific epistemic practices. These causal links still need to be established.

More recently, researchers have attempted to identify favourable conditions for “breakthrough” or exceptionally creative research. Hollingsworth and Hollingsworth [Hollingsworth Reference Hollingsworth and Hannaway2008; Hollingsworth and Hollingsworth Reference Hollingsworth and Hollingsworth2011] identified exceptional research in the life sciences and looked for common conditions under which this research took place. They concluded that “major discoveries tended to occur more frequently in organisational contexts that were relatively small and had high degrees of autonomy, flexibility, and the capacity to adapt rapidly to the fast pace of change in the global environment of science” [Hollingsworth Reference Hollingsworth and Hannaway2008: 321]. Intraorganisational conditions include moderately high scientific diversity (across the organisation and internalised in the scientists it recruits) as well as high intensities of communication and social integration [Hollingsworth and Hollingsworth Reference Hollingsworth and Hollingsworth2011: 17-40]. With a similar objective, Heinze et al. used a survey in which experts from human genetics and nanoscience/nanotechnology nominated more than 400 highly creative research accomplishments [Heinze et al. Reference Heinze, Shapira, Senker and Kuhlmann2007] and conducted 20 case studies of research groups in which such accomplishments occurred [Heinze et al. Reference Heinze, Shapira, Rogers and Senker2009]. They found “that creative accomplishments are associated with small group size, organisational contexts with sufficient access to a complementary variety of technical skills, stable research sponsorship, timely access to extra-mural skills and resources, and facilitating leadership” [Heinze et al. Reference Heinze, Shapira, Rogers and Senker2009: 610].

The findings of both studies are very general, mainly because the researchers looked for commonalities of all their cases in order to identify the most important conditions. The downside of this analytical strategy is that it does not significantly contribute to explanations because it cannot reveal what mechanisms produce specific outcomes under specific circumstances. This kind of explanation requires comparative case studies that systematically vary important conditions and outcomes, as has been attempted by Laudel and Gläser [Reference Laudel and Gläser2014]. The same idea shaped an internationally comparative study of the impact of changing authority relations on conditions for scientific innovations. The project started from the observation that the overall changes in the governance of research (see section 2. above) have altered authority relations concerning research content, and asked how these changes have modified the conditions for scientific innovations. The comparison of the development of innovations in physics [Bose-Einstein condensation, Laudel et al. Reference Laudel, Lettkemann, Ramuz, Wedlin, Woolley, Whitley and Gläser2014b], biology [evolutionary developmental biology, Laudel et al. Reference Laudel, Benninghoff, Lettkemann, Håkansson, Whitley and Gläser2014a], education research (international large-scale student assessments, Gläser et al. Reference Gläser, Aljets, Gorga, Hedemo, Håkansson, Laudel, Whitley and Gläser2014) and linguistics (computerised corpus linguistics, Engwall et al. Reference Engwall, Aljets, Hedmo, Ramuz, Whitley and Gläser2014) in four countries demonstrated the varied impact of changing authority relations [Whitley Reference Whitley, Whitley and Gläser2014]. In particular, it turned out that, while changes in authority relations might have led to more flexibility at the level of national science systems, many rigidities remained in place and new ones (such as the short-termism of funding) have emerged. The opportunities to develop innovation depends on the former’s epistemic characteristics, which translate into the innovation-specific necessary features of careers, resource allocation schemes, and evaluation practices [Whitley Reference Whitley, Whitley and Gläser2014].

3.2. The analysis of governance in constructivist studies of scientific research

The foundational constructivist studies saw practices of governance affecting the production of scientific knowledge but did not consider them because they did not seem to make a difference for the questions ethnographic observers wanted to answer at that time. Knorr-Cetina [1981: 68-93; 1982] observed the impact of governance—the constant need for scientist to adjust their research agenda to interests of other actors such as funding agencies—and concluded that instead of scientific communities, “transepistemic arenas” constitute the relevant context for researchers. Latour and Woolgar [1986 [1979]: 187-233] derived from their ethnography the model of a “cycle of credibility” that links the production of knowledge claims to the conditions under which the resources for this production are obtained. Fujimura’ [Reference Fujimura1987] showed that researchers construct “do-able” research problems, and that the “do-ability” of research problems does not only depend on epistemic factors but on conditions constructed by governance as well.

In spite of these observations, there are no systematic accounts of how (by what means and with what effects) the governance arrangements in which researchers are embedded, and to which they adapt, modify the content of the knowledge produced by those researchers. The sociology of scientific knowledge was (and still is) interested in different questions. The questions asked address a deeper level of researchers’ engagement with material objects, knowledge, or with each other. Governance, and its function of maintaining research processes, constitutes a background that is implicitly disregarded. There are degrees of disregard, however, which create a spectrum.

We describe this spectrum of considering governance with examples from more recent studies of research content. At one end, we find studies that look so deeply into the construction of scientific knowledge that they do not even see governance. For example, Hoeppe [Reference Hoeppe2014] conducts an ethnomethodological study to answer the question “How, then, do researchers achieve agreement on what constitutes a successful combination of data?” [ibid: 245]. He describes the institutional context and the governance of data production and sharing [ibid: 245-246] and proceeds to studying practices of “working data together”. Organisational conditions and governance completely disappear, and the practical problems faced by researchers are the only focus.

A next step from this pole towards increasing consideration of governance is the ethnography by Owen-Smith [Reference Owen-Smith2001], who studies the management of laboratory work and explores in detail the connection between social differentiation and practices of scientific discussions in the lab. Although the social differentiation is greatly influenced by funding (some researchers have their own funding, others depend on the director’s grants), the conditions for acquiring these grants and their impact on the directions of research in the laboratory are not discussed. Conditions shaped by governance are present as a background throughout the study but are not treated as relevant to the question asked.

Hackett [Reference Hackett2005] also discusses the functioning of research groups. He describes the role of a research group's “identity”, the way control is exercised in the lab, approaches to risk, competition, and funding. The outside world plays a far more important role in his account than in Owen-Smith’s because he considers the embeddedness of research groups in the competitive world of bioscience as an important condition. However, the conditions shaped by governance (competition, conditions under which grants are awarded, careers) are not considered with regard to their influence on research content. They form a background that obviously shapes behaviour but whose impact on research content remains obscure.

Kleinman [Reference Kleinman1998] explicitly argued that laboratory studies had neglected the embeddedness of laboratories in larger structures, and set out to explore the impact of such structures:

Kleinman showed that agrichemical companies defined the research agenda of the laboratory, and the processes through which this occurred. The laboratory worked in an applied agrichemical field, and the disease control agent it worked on was compared to commercial fungicides with regard to their cost-benefit ratio. He also described how the lab’s dependence on a commercially produced research material, Taq polymerase, created problems for research, and how difficult it would have been to circumvent buying the polymerase due to patent protection. Turning to the university administration and its role in patenting and licencing issues, he finally demonstrated the potential of this embeddedness of the lab to hinder research collaborations.

In his laboratory study, Kleinman identified three “pathways of impact” that make governance issues “reach through” to research content. He was not able to establish the extent to which they actually change research content because his ethnography was limited to showing that the research he observed proceeded as it did under the specific conditions created by governance. He could not consider the impact of alternative governance arrangements.

These laboratory studies mark a spectrum of the rising awareness of, and interest in, issues of governance. The collection of laboratory studies we perused is unevenly distributed across this spectrum. Most of the studies were at the “Hoeppe pole” of the spectrum, while only few were at the “Kleinman pole” (Tousignant Reference Tousignant2013 on the cessation of research due to a lack of resources in a Senegalese lab coming closest). We hasten to add that we do not judge the value of any of these studies. Our spectrum expresses the degree to which different kinds of research questions asked by laboratory studies are positioned to yield findings on the ways in which governance includes research content. We do not imply any normative statement about “right” and “wrong” or “important” and “unimportant” research questions.

3.3. Undone science

Another stream of research relevant to our question about the impact of governance on research content concerns knowledge that has not been produced. As the small but very rapidly growing body of literature on this topic indicates, this question is as complex as its dominant counterpart––the production of knowledge––and fraught with additional ontological, conceptual and methodological problems [Frickel Reference Frickel2014a; Reference Frickel, Kleinman and Moore2014b]. The topic is also inseparable from the impact of governance on research content because the notion of governance having an impact implies that different governance would produce different knowledge. Any governance that contributes to the existence of specific knowledge also contributes to the non-existence of other knowledge. This is why studies addressing our question explicitly or implicitly struggle with the problem of counterfactuals. What would have happened if the governance were different?

There are many forms of absent knowledge and many factors contributing to absences [Gross Reference Gross2007]. Theoretical and methodological traditions, the presence and absence of opportunities to produce specific knowledge and interests in the presence of some or the absence of other specific knowledge can all contribute to “undone science” [Frickel et al. Reference Frickel, Gibbon, Howard, Kempner, Ottinger and Hess2010]. In the context of our research question, we are interested in science that is undone because of governance. In these cases, the lack of specific knowledge is often pointed out by interested parties (Hess Reference Hess2007: 22, see also 2.6). These cases are very useful for the study of undone science because specific absences of knowledge are not easily identified beyond the trivial case of researchers framing their contributions by describing gaps in their community’s knowledge. Furthermore, we can limit our discussion to cases in which the absence of scientific knowledge is due to specific research practices that are shaped or maintained by governance.

Of the four case studies presented by Frickel et al. [Reference Frickel, Gibbon, Howard, Kempner, Ottinger and Hess2010], the “chlorine sunset controversy” is of particular interest to our question. The controversy addressed a regulatory paradigm that influenced the construction and articulation of research priorities. The regulatory paradigm required that each chlorinated chemical be individually tested for harmfulness. Opponents pointed out that the entire class of chlorinated chemicals was likely to be dangerous and demanded research that systematically addressed this problem. They lost, not least because the chemical industry had a strong interest in maintaining the traditional paradigm.

These observations resonate with Kleinman and Suryanarayanan’s [Reference Kleinman and Suryanarayanan2013] study of the “Colony Collapse Disorder”––a honey bee colony’s sudden loss in its adult population. The authors demonstrate that the dominant “toxicological epistemic form” underlying the regulatory paradigm for insecticides exclusively focused on lethal doses of individual insecticides for individual honey bees and thus “ignored-meaning that it failed to study, indeed could not study or would not consider seriously-possible evidence of the effects of low or ‘sublethal’ levels of insecticides” [ibid: 497-498]. Information provided by commercial beekeepers requires a different epistemic form for approaching the problem but is largely ignored. As in the case of chlorinated chemicals, a dominant epistemic form that is incapable of producing specific knowledge is maintained by governance arrangements for regulating a specific industry.

These two examples illustrate both the potential and the current limitations of studies of undone science. Such studies provide the epistemological opportunity to enhance our understanding of the ways in which governance shapes research content by investigating what happens in the “shadow” of governance (which does not imply that all this is necessarily unintentional). However, realising this potential requires overcoming the exclusive focus on consequences of undone science outside science and including the impact of undone science on the knowledge production of scientific communities.

3.4. Strength and weaknesses of influence-searching studies

This brief review of some recent research on the construction of scientific knowledge indicates a great but largely unused potential. The ways in which conditions of research are translated into changed knowledge have been studied in great detail by many scholars. Many of these studies also include a comparative assessment of influences in complex situations. They explore the content of knowledge production, often benefitting from a long and deep immersion in the field under study. sts has the research tradition and expertise that makes it possible to trace the impact of governance to its ends––the content of knowledge being produced by the governed research.^{Footnote 13}

From the point of view of our research question, it is a pity that sts has other interests than exploring the causal link between governance and research content. One possible reason for this reluctance may be the fear of a loss of “empirical resolution”: “If we choose a unit of analysis larger than the actual site of action, we remain removed from the indeterminacy which marks the situation” [Knorr-Cetina Reference Knorr-Cetina1981: 43]. Another reason is likely to be the dominant descriptive orientation of the sts mainstream [Frickel Reference Frickel2014a: 89], which is at odds with the inherently causal interest underlying this review, an interest we appear to share with the new political sociology of science [ibid.; Frickel and Moore Reference Frickel, Moore, Frickel and Moore2005b: 8-9].

Addressing causality in the impact of governance on research content would also raise two additional questions. First, such research would need to be comparative because assessing variations of governance and changes in knowledge content is necessary for causal ascription. It is very likely that governance has not surfaced more strongly in most ethnographies because they are single-case studies.Since no variation of governance can be observed in single-case studies, governance is treated as a negligible background condition. The necessity to systematically vary conditions created by governance in studies of knowledge construction points to a problem of research capacity. The necessary depth for exploring research content can only be reached for one case in projects conducted by just one researcher. Second, a comparative approach would necessarily include the comparison of research practices and research content across fields. This poses interesting challenges to sts [Laudel and Gläser Reference Laudel and Gläser2014; Gläser and Laudel Reference Gläser and Laudel2015b] but, again, sts is the field best equipped to meet them.