1. Introduction

An archive of around 1,000 e-mail messages from and to scientists at the Climatic Research Unit (CRU) at the University of East Anglia, together with related documents, was hacked and published without permission from the authors in November 2009. The correspondence offered an intimate and revealing picture of the practices involved in the construction of scientific knowledge and, more specifically, a record of discourse in a scientific community that felt under attack by critics and skeptics (Pearce Reference Reich2010; Ryghaug and Skj⊘lsvold Reference Shapere2010). In the antagonistic atmosphere in the weeks that followed, some writers argued that the correspondence showed the climate scientists to have engaged in unprofessional behavior of various forms: they alleged that the scientists had conspired to misrepresent the climate record to exaggerate warming trends, for example, and that they had colluded to suppress alternative hypotheses by influencing the peer review processes of scientific journals. Five authorities subsequently carried out inquiries into the professional conduct of the CRU scientists and of their colleagues at other institutions: their reports, published in 2010, largely exonerated the scientists of these charges. Nonetheless, the accusations and controversy underlined the contested and politicized nature of climate research (Lahsen Reference Leonelli, Howlett and Morgan2010).

A further important issue in the controversy, which has been less easy to dismiss, has been the demand for access to empirical data. Some critics of climate science had long believed that the raw data would reveal little evidence of anthropogenic climate change. They had increased their efforts in previous years to gain the release of data from the CRU, in many cases appealing to the Freedom of Information Act 2000 in the United Kingdom. Climate scientists had come to regard these applications as a campaign to waste their time and to gain material by which to undermine climate research unfairly (Heffernan Reference Heffernan2009; Mann Reference Merton and Storer2012, 200–201). Once the correspondence was in the public domain, the critics cited some messages as showing that the climate scientists had conspired to frustrate requests for access to data in order to prevent external scrutiny of their work. Many of these critics welcomed the hacking and publication of the private correspondence as retribution against the scientists for resisting earlier legitimate requests for transparency.

The reports of the subsequent five inquiries dwelled at length on the issue of access to empirical data. Some noted a reluctance of the climate scientists to place data in the public domain and commented on the desirability of sharing scientific data with other scientists and the general public.

For example, the report of the House of Commons Science and Technology Committee in the United Kingdom quoted the reply that Phil Jones of the CRU had sent to Warwick Hughes, who had asked for the raw data held by the CRU: “Even if WMO [World Meteorological Organization] agrees, I will still not pass on the data. We have 25 or so years invested in the work. Why should I make the data available to you, when your aim is to try and find something wrong with it.” The report commented, “On the face of it, this looks like an unreasonable response to a reasonable request. As Lord Lawson put it: ‘Ask any decent scientist and they will say the keystone for integrity in scientific research is full and transparent disclosure of data and methods’” (House of Commons Science and Technology Committee 2010, 12).

The report then summarized further arguments offered by Jones and other witnesses, which suggested that it was neither necessary nor possible for them to release all data: portions of the data were already available from other sources such as the Global Historical Climatology Network in the United States, the CRU was prevented from publishing some of the data by commercial agreements, most scientists preferred to work with adjusted data rather than with raw data anyway, and the CRU had no special duty to provide raw data to the general public. The committee appeared to accept some of these points and sympathized with the frustration that it suggested Jones must have felt in handling requests for data that he believed were motivated by a desire to undermine his work. Nonetheless, the report concluded, “In our view, CRU should have been more open with its raw data and followed the more open approach of NASA to making data available. … Transparency and accountability are of increasing importance to the public, so we recommend that the Government reviews the rules for the accessibility of data sets collected and analysed with UK public money” (House of Commons Science and Technology Committee 2010, 14; I have deleted an extraneous word “are”).

A few months after this British parliamentary report appeared, an inquiry panel established by the University of East Anglia and chaired by Sir Muir Russell published its report. It too dealt at length with the question of transparency and access to empirical data, finding that the CRU had been unhelpful in responding to requests (University of East Anglia Reference Kaye, Heeney, Hawkins, Vries and Boddington2010, 86–96).

2. Debates about Data Sharing

The discussion within the 2009 climate science e-mail controversy is a chapter in a continuing debate about two practices: sharing data with accredited researchers in the same or other scientific disciplines and making data available to the general public. In recent decades, some legal frameworks have evolved that allow members of the public without special credentials to request empirical data from research institutions and funding bodies. With this development, the modalities and practicalities of public access to empirical data have come to resemble more closely those of data sharing between researchers.

Meteorologists have one of the longest traditions of sharing and exchanging data, such as temperature and precipitation records. The International Meteorological Organization, established in 1873, assumed responsibility for coordinating the international standardization and exchange of weather data (Edwards Reference Edwards2010, 51–59). There is also a long history of data sharing in the empirical social sciences (Sieber Reference Suppes, Nagel, Suppes and Tarski1991). More recently, the practice of data sharing gained new impetus in genomics, especially from the Human Genome Project, the international effort that ran from 1990 to 2003 to map the human genes and make the information available for further biological study (Kaye et al. Reference Lahsen2009). Many bodies, including the Organization for Economic Cooperation and Development (Arzberger et al. Reference Arzberger, Schroeder, Beaulieu, Bowker, Casey, Laaksonen, Moorman, Uhlir and Wouters2004), have called for an international regulatory regime to promote access to data.

Proponents and opponents of data sharing put forward several arguments (Fienberg, Martin, and Straf Reference Fienberg, Martin and Straf1985). The arguments in favor of data sharing include these:

1. 1. Science depends on replication of results: science progresses most speedily and objectively if multiple teams can check one another's data and interpretations of them (Hones Reference Hones, Fine, Forbes and Wessels1990). Scrutiny may detect mistakes as well as fabrication of data.

2. 2. Notwithstanding the previous point, it is difficult in some cases to redo observations, especially of historical phenomena such as climate change. In these cases, the original data may be the only source of evidence, and any further work depends on sharing them.

3. 3. Considerations of efficiency and cost effectiveness suggest exploiting available empirical data as fully as possible, including reusing data in research projects other than the one in which they were gathered and for both scientific and commercial ends. If scientists were not allowed access to the empirical data of other researchers, wasteful duplication of effort would ensue.

4. 4. Science is governed by an ethos of unconditional sharing of all knowledge, perhaps akin to Robert K. Merton's norm of communalism (Merton Reference Morice, Kennedy, Rayner and Jones1942/1973). More specifically, ownership of data gathered in publicly funded research rests with the community, and scientists who carry out that research have no right to limit access to the data. Grant-giving agencies in several countries, acting partly on this argument, now impose a requirement to archive data in publicly accessible repositories as a condition of funding.

Arguments advanced for preventing dissemination of empirical data, by contrast, include these:

1. 5. Science is a competitive domain, and secrecy can offer a competitive advantage to individual researchers. A feeling of proprietorship with respect to empirical findings can spur scientists to greater efforts: this suggests that there may be a trade-off between transparency and scientific progress, especially in areas of intense research.

2. 6. Empirical data represent a form of intellectual property in which the scientists who have gathered the data have invested ideas and work: if they are not allowed some degree of (perhaps time-limited) exclusive use, scientists will be reluctant to make that investment. This motivation holds more strongly where there is a prospect of commercial exploitation of the data. In some cases, researchers are expected to make their data freely available, as in the construction of large-scale databases in astronomy: in those cases, other professional rewards must compensate for the loss of intellectual property (Finkbeiner Reference Finkbeiner2001).

3. 7. Confidentiality, privacy, and public security considerations weigh against releases of empirical data in particular cases (Bezuidenhout Reference Bezuidenhout2011).

One finds variants of the latter three arguments being put forward—explicitly or tacitly—also in industrial science, which holds to social norms different in part from those of academic science (Dasgupta and David Reference Dasgupta and David1994). There are reports of instances in which pharmaceutical firms have denied access to data to medical researchers, for example, even when the latter have been reporting their own research (Baty Reference Baty2005).

In some cases, arguments for and against the release of data clash. A recent example is offered by research by virologists in the Netherlands and the United States into an H5 influenza strain that, after genetic alteration, is easily transmissible between ferrets. Press reports raised public concern that the pathogen could be converted into a biological weapon. In the United States, the National Science Advisory Board for Biosecurity (NSABB) found that there was significant potential for harm in publishing the data and recommended that experimental results be withheld from publication (Berns et al. Reference Berns2012). The two research teams announced a 2-month moratorium (Fouchier et al. Reference Fouchier2012), but several other scientists called for timely publication of the full data notwithstanding the NSABB warning. For example, Peter Palese and Taia T. Wang, after arguing that the NSABB had overestimated the security risks, appealed to a widespread view of scientific practice, writing that “publishing relevant methods and data related to important experimental findings is fundamental to progress in science. Enabling others to repeat published data makes it possible to expand on what is known; the cycle of publication followed by experimental replication and advancement has resulted in the majority of our scientific and medical achievements. With specific reference to influenza viruses, the free flow of data has enabled the timely development of vaccines and other medications for seasonal and pandemic disease, and has, without question, saved millions of lives” (Reference Pearce2012, 2211).

3. Perspectives from Philosophy of Science

In the rest of this article, I focus on arguments 1–4 outlined above, advanced in favor of data sharing. Many scientists appear to accept and act on these arguments. For example, when the OPERA (Oscillation Project with Emulsion-tRacking Apparatus) collaboration at the Gran Sasso National Laboratory in Italy in 2011 announced that it had apparently timed neutrinos traveling faster than light from the CERN (European Organization for Nuclear Research) laboratory in Switzerland, the team made a public presentation and released its empirical data, inviting the neutrino physics community to search for sources of error that would account for the result (Reich Reference Ryghaug and Skj⊘lsvold2011). There have also been some experiences in sharing empirical data to harness large groups of researchers, in some cases amateurs, to tackle scientific problems (Nielsen Reference Palese and Wang2011).

By contrast, the stance, “Why should I make the data available to you, when your aim is to try and find something wrong with it,” expressed by Jones, seems at odds with Karl R. Popper's ideal of critical transparency and the disinterested effort to refute hypotheses. However, philosophy of science has also moved away from this picture of scientific practice in recent decades. This would not be the first time that discussions of scientific practice in the scientific community, in policy-making circles, and among the general public have been influenced by a view of science that professional philosophers of science regard as outdated and as no longer tenable.

From a philosophy of science perspective, there are several questionable aspects of arguments 1–4. First, the arguments seem to depend in part on the assumption that scientists need expend little effort to gather empirical data, which reach scientists in a way akin to messages sent by nature. If this were so, it would be plausible to portray data as a windfall that belonged to the scientific community as a whole. This view, however, holds only within the most heavily idealized theories of scientific practice. A more realistic view is that, virtually throughout modern science, data gathering requires a great expenditure of effort. This is particularly visible in precision research, such as research to determine the value of constants of nature and other physical parameters, where the reliability of a result depends on care expended to calibrate instruments and to control for sources of experimental error. In biomedical research, similarly, credibility depends on following agreed protocols conscientiously. In areas of big science, finally, such as high-energy physics, researchers heighten the credibility of their claims by emphasizing the effort that went into constructing their equipment: they often open conference talks by displaying pictures of impressive experimental installations, for example. The realization that gathering data is an effort-intensive activity blunts the force of argument 4 and strengthens the case for regarding data as a researcher's intellectual property.

Second, the arguments seem to depend in part on the assumption that there is a discrete phase of scientific practice in which scientists deal with what can be regarded as raw or unprocessed data. In contrast to this, it seems plausible to hold that the agency of the researcher is visible from the first moment at which data arise. The researcher exerts influence on what comes to be seen as the data, by selecting them. As studies of Robert A. Millikan's laboratory notebooks have been influential in suggesting, researchers routinely select some data points and reject others on a variety of criteria, which can be described as both cognitive and aesthetic (Franklin Reference Franklin1981; Brewer and Chinn Reference Brewer, Chinn, Hull, Forbes and Burian1994). The CRU correspondence, too, shows climate scientists employing scientifically legitimate techniques of processing and trimming data. This realization, too, lends weight to the idea that data are the output and thus the intellectual property of the researcher.

Third, the arguments seem to rest on the expectation that a second group of investigators can easily identify errors in the data collected by a first group. Several lines of thinking in philosophy of science suggest to the contrary that empirical data are a product of highly specific circumstances, in such a way that it is difficult for researchers not intimately involved in the collection of the data to understand their origin or to identify errors in them. This view of empirical data was suggested, for example, by James Bogen and James Woodward, who wrote that “many different sorts of causal factors play a role in the production of any given bit of data, and the characteristics of such items are heavily dependent on the peculiarities of the particular experimental design, detection device, or data-gathering procedures an investigator employs. Data are … idiosyncratic to particular experimental contexts, and typically cannot occur outside of those contexts” (Reference Bogen and Woodward1988, 317). The three considerations advanced up to now can be summarized in part by the thesis that the act of observation interacts with the external world to cocreate empirical data and that observational evidence is, therefore, theory laden (Shapere Reference Sieber1982; Heidelberger Reference Heidelberger and Radder2003).

Fourth, the arguments seem to presuppose that empirical data can be transferred from one setting to another without much difficulty. Philosophy of science has drawn attention to contextual aspects of science, especially to the idea that empirical data have only contextually dependent validity and are not easily transferable from one context to another. This could be partly a consequence of theory ladenness of evidence. It could also be due to the idiosyncratic nature of empirical data. Other work on biological and ecological data suggests that, using standardized databases, researchers can successfully transfer empirical data across research contexts, but knowledge of the local context in which data were gathered remains important to the reuse of data (Zimmerman Reference Zimmerman2008; Leonelli Reference Mann2010).

Fifth, the arguments are based on the assumption that empirical data directly play the role of evidence for and against scientific hypotheses. Bogen and Woodward (Reference Bogen and Woodward1988) suggest that, whereas this view was widely held earlier, it is in fact inadequate. There is no close relation between empirical data and hypotheses: instead, data are used as grounds for identifying phenomena—stable and reproducible causal modes in which reality is articulated—and the existence and properties of phenomena constitute evidence for and against hypotheses. If Bogen and Woodward are correct, empirical data are not directly employed in the processes of theory test.

These five considerations are drawn from theorizing in philosophy of science and, in the form in which I have expressed them, may seem too abstract to have a bearing on the concrete case of climate science controversies. I believe the opposite is the case, as the following discussion shows.

In climate science, there are three distinct things to which researchers commonly apply the name ‘data’. Each of these things plays some of the roles that we attribute to data in simplified accounts of scientific practice. First, there are temperature records and similar collections of raw observational data. These are unsystematic and patchy. They are unevenly spaced in place and time, and different portions use different instruments, rely on different assumptions, and are affected by different errors.

Part of the work of the CRU and of similar climate research centers is to process these records and convert them into a regular and gridded global temperature data set, such as HadCRUT (Morice et al. Reference Nielsen2012). Producing a data set such as HadCRUT from raw temperature records involves, for example, blending land station data with marine data to produce global coverage. It also involves, sometimes, replacing temperature readings with revised values and compensating for urbanization effects. These data sets are the second thing that climate scientists call ‘data’. They are the entities that are used to estimate trends in global mean temperature and to test climate models. In some respects, these entities can be regarded as an example of what philosophers of science call a ‘data model’, a mathematical representation of a body of empirical data (Suppes 1962). The function of a data model is not to reproduce the data points in full detail but to identify a significant component or pattern in the data.

The third thing that climate scientists call ‘data’ is even further removed from raw observation and is the outcome of modeling. As Paul N. Edwards writes, “Virtually everything we now call ‘global data’ is not simply collected; it is checked, filtered, interpreted, and integrated by computer models” (Reference Edwards2010, 188). These models include simulation models based on physical theory, reanalysis models constrained by weather observations, and data analysis models, which offer algorithmic compressions of important components of observational data (xv).

In this light, elements of the controversy about access to data in climate science are revealed as partly misguided. The burden of some of the complaints of critics and skeptics of climate science was that scientists at the CRU and other institutions were reluctant to provide the neutral evidential base on which they had constructed hypotheses about anthropogenic climate change—the ‘data’, in other words, on which the scientists’ ‘models’ were (erroneously, perhaps) based. If we are justified in thinking that three distinct things play the roles of data in climate science, then it is possible to argue that the climate scientists had complied all along with the requirement to allow access to data. After all, global temperature data sets, such as HadCRUT, as well as discussion of the assumptions used in producing such data sets were in the public domain; the same holds for the results of modeling that also go under the name ‘data’. The critics could still claim that the raw observational data had not been published in full, but it appears that such data play a much smaller role in grounding our knowledge of the behavior of the global climate system than one might initially think. The empirical base is inextricably bound up with the interpretation of it in the form of models.

4. Conclusion

Transparency in science is a laudable aim for both epistemic and social reasons, and the ideal of shared ownership and use of empirical data and other forms of output of science speaks to our highest values. There seems little doubt that, in practice, climate scientists at the CRU and other institutions in the years before 2009 could have been more forthcoming in placing material in the public domain. However, one of the social roles of philosophy of science is to complicate and problematize issues relating to scientific practice where we believe that to be warranted. Recent research in philosophy of science suggests that traditional notions of empirical data are, to some extent, an idealization. Those who look to the category of empirical data as a perfectly objective token from nature, unspoiled by human interference, that speaks for itself and is able to resolve scientific controversies are bound to be disappointed. This outcome does not cast the objectivity of science or of particular scientific findings into doubt, but it means that we must expend more effort in describing the way in which this objectivity arises from the human engagement with nature.