AIP STUDY OF MULTI-INSTITUTIONAL COLLABORATIONS

I. INTRODUCTION
[Table of Contents]

The analysis is based on 165 interviews on 19 high-energy physics experiments approved 1973-1984 at Brook haven National Laboratory, Fermi National Accelerator Laboratory, and Stanford Linear Accelerator Center. In most cases, interviews were conducted by project staff in the interviewee's office using standardized question sets for each interviewee category: physicists, graduate students, and engineers and technicians.

II. GENERAL ORGANIZATION AND MANAGEMENT
[Table of Contents]

A. Social Origins of Collaborations
[Table of Contents]

The period covered by our sample began two years after Fermilab became operational, one year after the SPEAR storage ring at SLAC became operational, and included the inauguration of the Fermilab tevatron, the Cornell Electron Storage Ring, and the PEP storage ring at SLAC. The impact of these accelerator developments on the formation of collaborations was dramatic. Five of the seven collaborations that performed the Fermilab experiments in our sample came together to design experiments for Fermilab's opening; a sixth was a response to the tevatron. Three of the six collaborations that performed SLAC experiments came together to plan PEP experiments. And the CLEO collaboration formed to build a detector for the Cornell storage ring. Even among the experiments not linked to a major accelerator innovation, there were two whose collaborations formed to take advantage of innovations in beamline or targets. While it would be an oversimplification to label the formation of collaborations as "accelerator driven," because some of these experiments also aimed to produce novel detector designs, to test particular theories, or to resolve experimental controversies, the implication of this sample is that the construction of a new accelerator is a sure way to stimulate a shake-up in the pattern of working relations among high-energy physicists.

Experimenters discovered their common interests, or at least the willingness to work together in a collaboration, through a number of channels. Friendships, not surprisingly, were often important to collaboration organizers, and there is evidence of people who met as far back as college trying to find their way into doing an experiment together. But between the sheer number of people needed to mount even the smaller experiments in our sample, the vagaries of individuals' schedules and commitments, and the need to tap particular forms of expertise, only two collaborations were successfully assembled entirely out of the pre-existing professional relationships and personal contacts among the organizing collaborators. The rest required organizers to go beyond their circle of physics friends. The organizers of an under-staffed proto-collaboration sought others through their grapevines and by buttonholing colleagues at conferences or seminars, but other, more formal means were also employed. There are two instances in our sample of collaboration organizers approaching physicists they had never met on the strength of those physicists' published accomplishments. Summer studies to discuss possible PEP experiments spawned both PEP and SPEAR experiments. Three proto-collaborations held open meetings where potentially interested physicists could come to discuss the proposed experiment and meet the organizers. And five experiments were performed by "shot-gun marriages"—collaborations that laboratory administrators had brokered after receiving multiple proposals to use or build similar apparatus.

Interviewees were reluctant to talk about the reasons experimenters do not join collaborations, even when not required to identify the individuals or institutions involved. But two factors that blocked consummation of a collaboration were apparent. First was the perception that a physicist or his institution were over-extended and unwilling to cut back on other activities; no proponent of an experiment wanted its fate to depend on the contribution of someone whose best efforts were going elsewhere. Second was the difficulty an experiment's proponent could have within his home institution. There was one instance in our sample of a university physicist being unable to participate in an experiment he helped to design, because he could not convince his departmental colleagues, who administered the research funds, that support of his experiment made sense from a departmental perspective.

There is no apparent formula for picking collaborators successfully. Our sample included one collaboration where a falling-out between long-standing friends infected physics discussions involving their students; the members of that collaboration have not tried to work together again. The sample also included two instances of strangers becoming fast friends and long-standing collaborators. It does seem generally to be the case that brokered collaborations did not endure beyond the experiment that brought the collaborators together. That does not mean that brokering was a bad practice. If an experiment was attractive enough for multiple proto-collaborations to propose it, but the proposers were either unable to find each other on their own or unreasonably optimistic about their abilities to perform the experiment on their own, a brokered collaboration was the only way to get the experiment done. Rather, the simplest criterion for an experiment's social success—did the experimenters continue to work together?—should be modified for brokered collaborations.

B. Size and Composition of Collaborations
[Table of Contents]

The organizers of experiments understood that they needed to attract a vaguely-defined critical mass of physicists to an experiment in order to appear credible to administrators whose duty it was to assess the feasibility and desirability of a proposal. And a collaboration's leaders then had to deploy their resources so as to succeed in building, running, and analyzing data from the proposed experiment. These necessities pressed physicists to form collaborations to match the scale of their experiments rather than the norms previous experimentation; one experiment organizer intended to create an experiment for five to ten physicists but saw "the physics necessity" drive the experiment up to 40 collaborators. However, these necessities did not induce physicists to "pad" proposals in order to have a margin of safety should some part of the experiment prove unexpectedly taxing. To the contrary, interviewees usually described experiments as understaffed, shoestring operations, even when the collaboration seemed unusually large. For example, one interviewee, a post-doc on the experiment being studied, recalled feeling shocked that this experiment had nearly three times as many physicists as his thesis experiment, yet he then found himself terribly overworked during the experiment's construction.

Such individual impressions appear to have a collective reality: in five of the experiments in our sample, collaborations had to add institutions after the proposal had been approved in order to carry through the experiment; in three others, which were second or higher in a string of experiments, the collaboration added one or more new groups prior to the proposal in recognition of the difficulty in carrying on with their current complement; and in only one of the five brokered collaborations did the merging of proposals require the collaborating institutions to pare down their personnel.

Competition for the field's limited funds and personnel was an obvious factor in keeping collaborations lean in relation to the tasks they undertook. Our sample included one experiment that was successfully redesigned to require less material and people in response to the need to be competitive with other proposals vying for an empty experiment site at the accelerator laboratory. A quantitative study analyzing, over time, the number of proposals, the level of funding, the number of high-energy physics groups, and the availability of experiment sites could perhaps assess the importance of competition in keeping down the size of collaborations. However, the overall, qualitative impression from the interviews was that the understaffing of experiments was a cultural norm that many American physicists liked. This impression is strengthened by the dearth of similar impressions in the interviews of European physicists.

Understaffing had at least four advantages. First, it reduced resistance to tapping the knowledge and experience of those outside the collaboration (the "not-invented-here" syndrome). For example, our study of subcontracting indicates that LeCroy Electronics has successfully made a business of producing electronics modules that could be sold to many different experiments, and in one of the experiments in our sample, a physicist and engineer canvassed manufacturers to "learn what was available" for the processing of plastics that had desirable properties for the experiment.

Second, understaffing made it easy for every collaborating institution to be responsible for an important construction project and preempted worries about whether an experiment would address enough physics issues to satisfy the collaborators' needs for distinctive accomplishments. Almost absent from the interviews are stories of "turf disputes" with respect to either apparatus building or data analysis. There was only one experiment where the need for more personnel to build the apparatus induced among the graduate students a fear that there would be an inadequate number of thesis topics; here, a collaboration retreat for physics discussions yielded more than double the number of possible topics for the existing population of students.

Third, understaffing inhibited collaborators from hiding or underestimating problems because nobody could afford to be unreliable in an operation on which others were working so hard. At least among the collaborations of our sample, collaborators were sufficiently open with each other about their difficulties, or lack thereof, that there were several occasions in which collaborations profitably redeployed their resources in order to address areas that turned out to be more troublesome than initially expected. And fourth, understaffing limited the spread of credit for the making of an illuminating discovery. During an experiment's construction, one junior faculty collaborator recalled overhearing a conversation among the experiment's "gurus" over how they would divide the Nobel Prize they expected to receive for the experiment's results.

A minority of interviewees worried about the effects of doing experiments on a shoestring. The most common complaint was that graduate students spent time on intellectually unchallenging construction tasks when they could have been preparing and comparing various simulation and analysis programs. One interviewee feared that physicists were undertaking construction tasks which they were unqualified to do well and safely because of the budgetary impact of hiring skilled laborers.

Even though collaboration organizers kept collaborations as small as possible, the larger experiments created administrative positions within collaborations. Executive committees sought to insure that evolving component designs did not undermine collective coherence, to reallocate resources should the development of a particular component run into snags, and to provide the several senior members of the collaboration with a formal forum for discussing and reaching decisions on the collaboration's internal and external administrative problems. Software coordinators regulated changes to codes and programs used widely within a collaboration. Deputy spokespersons oversaw construction or data runs. And ad hoc publications committees served as referees between the producers of a physics analysis and a collaboration as a whole. Issues that were either the stuff of collaboration meetings or within the purview of a senior collaborator's unilateral actions in smaller collaborations were thus initially thrashed out in formalized subgroups of larger collaborations. But even for most large experiments, meetings of the entire collaboration remained the forum for making most decisions on basic strategies for designing, running, and producing results from the experiment.

One physicist, who worked on one of the smaller, early experiments from this study, left high-energy physics rather than work in ever larger collaborations. Some who have managed to keep working on smaller experiments looked at the larger collaborations with a mixture of perplexity and disdain. But the people on the inside found satisfaction in large collaborations even when they had expected to feel uncomfortable. One experiment organizer, who became a physicist to avoid becoming like his father, the president of a medium-sized company, still felt "pleasantly surprised" at the manageability of the personal interactions of a 40-physicist experiment, and succeeded in understanding and contributing to the experiment from top to bottom. He feared that he will not be pleasantly surprised again, but he was not leaving the field. Another interviewee found that the worst is over, for the moment, for high-energy physicists—that while the experimenters in the larger collaborations from our sample struggled to operate as a unit, his "super-large collaboration" at CERN forgoes unitary operations as "too unwieldy" and gives much more autonomy to his institutional unit, "which is a very manageable size."

Experiment organizers tended to worry about getting enough collaborators for an experiment rather than putting together a complementary blend of skills and sub-specialties. One organizer remembered welcoming a university group into a collaboration because "at that time we needed to have bodies." Another stated that balancing skills in building a collaboration was meaningless because everyone typically had to learn plenty of new things in the course of doing the experiment; to flesh out his collaboration, he looked for people with "a good reputation, solid people" who would inspire "mutual confidence" among all the collaborators. Only three collaborations (aside from two Soviet-American cases discussed below) came together as a result of complementary expertise among the participants, and of these three, one was an atypical experiment, poorly suited to graduate education, in that it aimed to make only one measurement in order to search for one hypothetical particle.

The overall impression from our interviews was that individual American physicists were presumed to be familiar with, if not expert in, all phases of an experiment, and a university group encompassed all the skills needed for an experiment. Throughout our sample, physicists invariably wrote and managed experiments' computer programs; though a few physicists were known as "computer-oriented" and specialized in working on data acquisition systems for experiments, they tended to be employed at national laboratories. Some physicists saw themselves or certain of their colleagues as primarily apparatus builders or data analyzers, but everyone wanted graduate students to participate in both types of work in the course of their thesis experiments. The one meaningful distinction among groups during the years covered by our sample was between those specializing in electronic and bubble chamber experiments. One electronic experiment was especially attractive to physicists in bubble-chamber groups who wished to switch or expand their groups' style of work.

The two Soviet-American collaborations in our sample both had their origins in particular experimental techniques that the Soviet groups had developed. Both Soviet and American interviewees commented on the conditions that created this grounds for collaboration. One Soviet interviewee reported he was labelled a physicist in the U.K. and the U.S.A., though he was considered an engineer in the U.S.S.R.; one of his American collaborators came away with the impression that individual Soviets specialize more permanently with the consequence that they become very adept at what they do and less able to keep up with experimental innovations. An American participant from a different U.S.-U.S.S.R. collaboration believed the Soviets have created a class of people "between what we would call engineers and physicists;" these people, though not interested in the physics results, did take data shifts and therefore were included as authors for experimental papers. The apparent Soviet practice of building groups out of experts in the technologies of experimental techniques as well as experts in the physics of elementary particles was perhaps due to the placement of Soviet researchers in Institutes of the Academy of Sciences rather than in universities. However, one Soviet interviewee also pointed out that cultivation of novel experimental techniques was essential to Soviet participation in international collaborations outside the U.S.S.R. Because the ruble was worthless outside the U.S.S.R., Soviet physicists could not bring money to a foreign experiment and therefore had to bring expertise and equipment that was not readily purchased in the west.

C. Multi-Institutionality
[Table of Contents]

During the period covered by our study, two types of institutional groups participated in high-energy physics experiments: research groups at the accelerator laboratories and university groups. Because academic high-energy physicists in the U.S.A. were funded as university groups, whose size and budgets were limited by both university and governmental dynamics, and because accelerator-based groups were fewer in number, collaborations could become larger only by including more academic institutions. The addition of institutions brought collaborations both extra resources and organizational complexity. This fundamental trade-off was probably the greatest source of daily friction within collaborations.

When a new institution was added to a collaboration, not only did the collaboration gain more physicists, it also potentially gained access to the institution's research and development laboratories, machine shops, computers, and an inexpensive supply of undergraduate labor. In the competition to join collaborations with experiments that are likely to be approved, physicists at universities with strong resources had an obvious advantage over those at universities with fewer resources, and the physicists at the accelerator laboratories were best situated of all. 12 of the 19 selected accelerator experiments had in-house collaborators, and university physicists often spoke of the desirability of access to the computers and machine shops of the accelerator laboratory at which they wished to do an experiment. Furthermore, inclusion of research groups at the accelerator laboratories may have actually reduced organizational complexity given the relative ease with which in-house physicists oversaw the installation of apparatus. (Three of the six collaborations without in-house collaborators included groups located within commuting distance of the accelerator and a fourth tried but failed to interest accelerator-based physicists in its experiment.) However, university groups during the period covered by our sample were neither impotent in the face of a lack of collaborators at the accelerator nor necessarily dominated by the accelerator-based groups with which they collaborated. Four of the six experiments without in-house collaborators did build large detectors, and there was one instance of a collaboration between university and accelerator laboratory groups where the university machine shops took on work for the laboratory physicists, who could not afford to pay their own shop to fabricate the desired components.

The collaborations in our sample took any of three approaches to their multi-institutionality and did not necessarily adhere to any one approach throughout the experiments they performed. Sometimes collaborations sought to take advantage of their multi-institutionality by making it a framework for their working relations. Sometimes collaborations ignored their multi-institutionality as irrelevant or ill-suited to the tasks they faced. And sometimes collaborations found multi-institutionality created outright liabilities that had to be lived with for the sake of getting the experiment done.

Collaborations tried to make a virtue of multi-institutionality when physicists wished to divide labor for efficiency or duplicate labor to insure reliability of results. The former was frequently employed by collaborations that built large, multi-component detectors that were considerable research and development projects in and of themselves. In 12 of the 19 experiments, most of the collaborating groups worked independently on components once the collaborations had set basic design parameters. When such management worked well, interviewees remembered healthy, productive intra-collaboration competition: each group sought to gain credit for building the best component, and to avoid becoming known as the builder of the component that limited the quality of the experiment's measurements. Such management entailed risks and difficulties as well. There were instances where the independently built components, once assembled at the laboratory, produced systems effects that made for extra work for some collaborators, although our sample of successful experiments included no examples of a component being substantially rebuilt in response to systems effects. There were also instances where one group's component caused or threatened to cause a setback to the collaboration's timetable, although our sample included only one example of an individual physicist seeming sufficiently blameworthy to leave the collaboration.

Systematic, institutionalized duplication of labor, as opposed to ad hoc double-checking of important or unexpected results, was much less common, presumably because physicists who were so uncertain about each other that they desired duplication usually did not collaborate in the first place. Only four collaborations tried duplicating labor; in three of them, one of the duplicating lines of research collapsed, and in two of these three, the efforts at duplication caused hard feelings within the collaboration. However, in the one collaboration that kept up the practice, physicists recalled the intra-collaboration competition as a "healthy tension;" the "really acrimonious" disputes that arose when the two institutions reached conflicting results were honorably intellectual, "and we always went out and drank beer after the big fights."

Most collaborations sought to ignore their multi-institutionality during data analysis, even if they exploited it while building apparatus. Their usual strategy was to concentrate data analyzers at the site of the best available computer for "crunching" the data and to base individual analyses on collectively produced data summary tapes and programs for reconstructing events in the detector. Seven collaborations also ignored their multi-institutionality while building their experiments. The most common precondition to this practice was that the experiment's technical drive stemmed not from someone conceiving of an improvement in detector design or strategy, but rather from academic or foreign physicists conceiving of an innovation in target or beamline and needing to work out their ideas in close coordination with collaborators at the accelerator laboratory. In these cases, the important technical divisions in the experiments did not coincide with the institutional divisions in the collaborations, and the collaborations assaulted their multi-institutionality via telephone, telex, e-mail, or meetings. In certain collaborations, multi-institutionality created outright liabilities that became apparent after it was too late for corrective action. In two cases, the needs of junior faculty to use an experiment to gain tenure within their home institutions induced behavior that other collaborators found objectionable. In one of these cases, a junior faculty member tried to set up a data-analysis chain based at his institution—presumably to demonstrate his competence in that phase of the experiment and to provide a basis for claiming personal credit for discoveries or important measurements—when the rest of the collaboration preferred to work collectively within the framework set up by a collaborator who was especially skillful in programming and analysis. The junior faculty member's collaborators found him secretive and his efforts counter-productive. In the other of these cases, the collaboration made two junior faculty at different institutions co-spokespersons, in order to avoid granting one the advantage of the title and role in his campaign for tenure, only to find that they competed to impose their personalities on the experiment and could not resist questioning and seeking to overturn each other's decisions. This collaboration found their behavior polarizing and a source of administrative confusion. Finally, there was one example where a disagreement between two physicists over the physics of the experiment took on an institutional cast in the eyes of one of the physicists' graduate students, who felt that the other physicist was unreasonably dedicated to finding objections to his analysis. Had the two physicists and student all been in one institution, the student could have appealed to a department chairperson or dean to mediate, but within the collaboration there was nobody with effective power over the physicists to mediate, and the student had no recourse to persevering in the face of unconstructive criticism.

Those collaborations in our sample that found multi-institutionality problematic all lived to publish results. As one of the co-spokespersons in the two-spokesperson experiment retrospectively realized, the collaboration contained plenty of "hungry people" who were not going to let problems between the spokespersons deprive them of data and publications, and a collaboration of bosom buddies leading "a bunch of bums" would have done far worse. However, the evidence here did suggest that collaboration organizers would be wise to consider how they can satisfy the needs of junior faculty in advance of committing themselves to including particular institutions, and that all post-doctoral physicists should resist, as professionally unethical, the temptation to involve their students in their intra-collaboration disagreements over physics.

D. Hierarchies and Social Relations
[Table of Contents]

Everyone in a collaboration knew who was a full professor, an assistant professor, a postdoc, a graduate student, an engineer, and a technician, but privilege of rank was rarely in itself the cause of strained social relations within collaborations. There was a dearth of stories about collaborations in which individuals inappropriately "pulled rank" on their juniors or lacked appropriate respect for their seniors. Occasionally students or postdocs claimed that senior people were making "the decisions that matter," but did not specify what the issues were. However, most physicists remember their collaborations as participatory democracies where decisions were made by consensus and where individuals were distinguished by knowledge, ability, and sometimes the degree to which they had invested their careers in the experiment. Engineers reported that they avoided collaboration meetings, which they felt turned into gripe sessions where "the physicists put you on the hot seat." But they professed comfort in dealing with physicists on a one-on-one basis. Graduate students and engineers enjoyed a symbiotic relationship in an experiment; the students kept the engineers posted on discussions in collaboration meetings and the engineers taught the students "hands-on science," which could prove professionally useful during slack periods in the market for high-energy physicists.

What strained social relations in our sample of collaborations more than the simple span of ranks they contained, were structural problems imposed by their size, their inability to reward and discipline those faculty-level members who could operate autonomously within their home institutions, and the pressures of competing with other collaborations. These factors sometimes had unequal impacts on collaborators on the basis of their rank or institution. But they were never cause for a physicist to leave an experiment in progress.

Six of the more recent collaborations in our sample bowed to the complexities of coping with their larger sizes and created executive committees that took care of some portion of the collaboration's business outside of collaboration-wide meetings. In five of these six instances, the collaboration meetings, which continued to be run as participatory democracies, remained the most important forums for discussions of the experiments' progress and results. Physicists not on the executive committees of these collaborations more or less cheerfully accepted that the committees were more effective or efficient than collaboration meetings at keeping the several groups coordinated, making decisions or reaching compromises on issues for which there was no collaboration-wide consensus, setting forth a collaboration "party line" in policy struggles with laboratory administrators or industrial suppliers, or just giving strong-minded senior people a more intimate setting in which to argue with each other. Junior people on these experiments retained the power to manage the work they were undertaking. In one of these five collaborations, when the graduate students felt the executive committee was overstepping its bounds in considering ways to divide up possible dissertation topics, the students themselves met and decided on a division of topics. The students' division was the one that was followed.

In the sixth case, interviewees viewed the executive committee's meetings as more important than the collaboration meetings, and physicists at opposite ends of the hierarchy were all initially unhappy with the collaboration's administration. The spokesperson felt the executive committee members were too parochial in guarding the privileges of and seeking credit for their home institutions, and insufficiently sensitive to the needs of the experiment as a whole; a graduate student felt the committee spent too much time debating who would present results that had not yet been achieved and too little time on the problems of technical coordination. As the collaboration neared completion of assembling the experiment's hardware, the graduate students were upset by uncertainty over the experiment's potential to generate enough dissertation topics. In contrast to the collaboration discussed above, the students were not able to resolve the issue on their own. The collaboration saved itself by holding a retreat; after several days of small-group and collaboration-wide discussion, all were convinced that the experiment could generate double the number of needed dissertation topics.

The collaborations in our sample lacked the administrative powers to reward and discipline their faculty-level members. The collaborations could not grant or withhold promotions, raise or lower pay, increase or decrease the number of students or postdocs a faculty member could hire, or provide quicker or slower access to a machine shop or research and development laboratory. All of these powers rested with the several institutions that employed the collaborators. Collaborations' lack of powers became problematic in three kinds of situations.

First, when a narrowly focussed experiment involved many more faculty than it had physics topics to address, the collaboration was prone to divisive disputes over the paucity of rewards available. In one experiment in particular, where the physicists designed and built a detector to search for a hypothetical process, collaborators became sensitive over who received credit for the narrow range of results all wished to produce, and collaboration meetings featured rancorous arguments over who would present results at upcoming conferences. This particular collaboration enjoyed a long life by investigating many ways the hypothetical process could occur and by serendipitously discovering another use for the detector. Other collaborations that performed specific search experiments disbanded after they had achieved their designated results, leaving some of their faculty members mildly disgruntled with their personal "return" on their investment of time and effort.

Second, when experiments could address multiple topics but only one topic at a time, or could yield data susceptible to multiple treatments in data analysis, collaborations were "Balkanized," and the inability of collaborations to discipline faculty led in some cases to easy dissolution. Individuals or small cliques of faculty would usually adopt one of the possible lines of inquiry and then fight over whose interests deserved collaboration-wide support. Although present in the upsilon string of fixed-target experiments (See Probe Report), this phenomenon was especially apparent in collider experiments. Electron-positron annihilations create only those particles whose masses equal the energy of the collision; thus collaborations comprised of physicists interested in particles of different masses always faced a contentious issue in deciding at what energy the accelerator should be set. Interviewees from one experiment in particular recalled "endless meetings" and "semi-annual blood lettings" when the collaboration had to take a position on what energy the collider should be run at and thus what kinds of events the collaboration could study. So long as the collider ran well and the data "pie" was growing, the collaborators could at least tolerate, if not enjoy, arguing over the size of the data "slices" and which data runs should be dedicated to whose favored accelerator setting. When the data grew harder to come by and the collection of a meaningful amount of data on some topics effectively meant no further data would be collected on others, the collaboration could not discipline its members to hold to a subset of physics goals that was consistent with what the accelerator was likely to deliver. Instead, the collaboration disbanded so that the faculty members could give more attention to other experiments, even though collaborators believed there were still data worth collecting.

Third, collaborations' inability to reward or discipline collaborators preempted or inhibited collaboration-wide debate over the design of detector upgrades. In our sample, the separate institutions applied for funding and controlled access to their laboratories and machine shops, and in two instances in our sample, that power was exercised independently of the collaboration. One collaboration ended debate over two competing designs for upgrading a particular component when one of the proponents of one of the designs succeeded in obtaining funding to build his design. In another collaboration, the original builder of a component had to accept a consultant's role to another institution's physicists in the building of an upgraded version because he could not commit his university's resources to building the upgraded version. (Although not noted by any interviewee, this autonomy of constituent groups within collaborations may limit the scope of collaboration democracy; long-range planning is removed from the arena of collaboration meetings, where physicists of all ranks may participate, and reverts to the arena of discussions over funding, where senior physicists typically predominate.)

The pressures from inter-experiment competition became divisive within three collaborations where one or some collaborators presented claims to have made a discovery or a difficult measurement. The delicate art, which physicists have long practiced individually, of balancing the fear of appearing foolish for publishing an erroneous claim against the fear of losing credit should a competitor publish a correct claim first was not readily practiced in these collaborative settings. In one case, the group leaders, convinced of the reality of a discovery and fearful of being "scooped," together drafted publications without giving others a meaningful opportunity to make criticisms or suggestions. They effectively dared the rest of their collaborators to remove their names from the author list of papers and suppressed debate that could have "fine-tuned" the paper to reflect more accurately a collective assessment of the certainty of the claims and the weak points of the experiment. In another case, a collaboration yielded to the enthusiasm of the proponents of a discovery claim when the accelerator laboratory's administration declined to provide beamtime to double-check the claim in advance of its public dissemination. The claim turned out to be erroneous, and the non-proponents now speak of the lack of appropriate skepticism and openness among the claim's proponents. In a third case, a disagreement between two collaborators, who were apparently struggling with a personal falling-out as well as their physics differences, blocked prompt publication of results. A third collaborator, who expressed dissatisfaction with his home institution and was presumably eager for a success that would gain him some leverage in his efforts to improve his situation, was disgusted that others published their measurements first while his collaboration struggled to draft papers to which all collaborators would subscribe.

There were no examples in our sample of an institution quitting a collaboration or being driven out of a collaboration because of destructive social dynamics, and there was only one instance of an individual physicist leaving an experiment in progress because of grievances with his collaborators. That incident did not involve structural, collaboration-wide strains, but stemmed from the physicist having lost the confidence of his collaborators in his ability to carry out the task for which he had taken responsibility. Institutions have left collaborations when they judged an experiment had reached the point of diminishing returns in comparison to their other options, and individuals often left collaborations when their institutional affiliations shifted. But virtually never did an institution or individual leave an experiment "in a huff" over treatment received. The overall impression from our sample is that physicists tolerated inequities and conflicts in order to be part of an experiment.

III. FUNDING
[Table of Contents]

The two governmental funders of high-energy physics in the U.S.A., the Department of Energy, which is a direct descendant of the Atomic Energy Commission, and the National Science Foundation, came into existence shortly after World War II and established traditions of funding academic researchers through their universities. While high-energy physics experiments have long since outgrown the capabilities of individual universities, university units continued to be administrators of research funds for their high-energy physicists. This arrangement made the cost of individual experiments difficult to calculate and compare, because a university's contribution was embedded in the cost of all the activities supported by its contract for high-energy physics research. However, there were at least two powerful reasons, besides institutional inertia, for this fragmented administrative framework: collaborations have been transitory while universities have been stable fixtures in the institutional landscape; and university groups have needed the potential to regulate the activities of individual faculty in the interests of maintaining a mix of activities that best served the needs of the department and its graduate students. Several interviewees had the impression that the funding agencies are increasingly insisting that experimenters provide a unified proposal for their consideration. Rather than funding an experiment through the support of several institutions' proposals, the agencies used the accelerator laboratories as central money-managers of funds for an experiment with several institutional collaborators. That was how the experiments on the PEP accelerator at SLAC were handled. This collaboration-centered study, however, is not a good foundation for generalizing on the evolution of agency practices.

Most of the experiments in our sample were funded through universities; this policy encouraged the multi-institutionality and the internationalization of collaborations. Interviewees sensed there were imprecisely defined limits, which were best left imprecisely defined, on how much money would be spent on the research of any single high-energy physics group. Thus any group with the ambition to design and build an expensive experiment, or to stretch an extant experiment to make more measurements than the group could handle, had to convince physicists from other institutions or countries to dedicate some portion of their institutions' resources to the experiment. (Research groups of the national laboratories were an exception to this statement; they also had the option of building expensive apparatus as a "facility" that would be available to other high-energy physics groups, and our sample included two experiments that used such facilities.)

Because our sample consists only of experiments that received funding and ran, one would not expect to find much evidence of funding procedures breeding resentment. And, in general, one would not have expected funding procedures to breed resentment because attracting collaborators (and their resources) and establishing scientific merit were neatly in tandem. However, occasional funding-based problems did crop up for the experiments in our sample. There was one apparent case of conflict over how closely theory and experiment must be related in order for an experiment to be worth funding; one group of experimentalists claimed that a similar proposal to the one selected for study here had been denied funding, but that a second version received favorable, generous treatment because theorists had subsequently developed a basis for predicting what results the experiment should achieve. In one collaboration between sets of relatively cash-rich and labor-rich institutions, responsibilities had to be carefully divided so as to give the labor-rich institution enough work to hold its interest without undermining the confidence of the cash-rich, who would have preferred to buy commercially available apparatus. In two collaborations where all but one of the institutions was DOE supported, members of the NSF-supported institution felt they had to be made responsible for a distinctive, clearly differentiated part of the experiment in order to gain the necessary NSF support to participate in the experiment. And finally, in another instance, one collaboration diversified its funding sources to the point where both the high-energy physics and the nuclear physics program offices of both DOE and NSF were involved; during the time it took to get the commitments from all four offices straightened out, the administrations of some of the involved universities loaned funds to their physics groups to begin work.

University-based experimentalists took for granted that their contracts with the funding agencies would cover travel, the support of post-docs and graduate students, and the operation of any university laboratories or shop facilities dedicated to the high-energy physics group. Uncertainty existed over the prospect of acquiring funds to buy the materials and services needed to construct major, new detector components. The uncertainty placed a premium on reusing equipment, and six experiments in our sample did not require much U.S. capital. Some experimentalists found virtue in this necessity and took delight in bypassing the extra scrutiny that came with requests for capital funds, short-circuiting the long lead-times associated with building new components, and quickly mounting experiments with previously built apparatus plus what could be built within the confines of the secure budget.

The prospect of applying for capital funds could not be indefinitely postponed, and university units took a variety of approaches to managing their members' needs to build major pieces of apparatus. At one extreme was a group whose members "don't really know what each other is doing" and relied on the low probability that several members would want to build apparatus in a given year. In the middle were groups with sub-groups that aggregated some of the members' needs, and groups whose members worked independently but appointed a Principal Investigator with the authority to work out an intra-group distribution of the resources provided by the government. At the other extreme were groups that operated as a unit, collectively deciding on the experiments they would or would not pursue.

IV. EXPERIMENT STRINGS, DETECTOR DEVELOPMENT, AND NEW TECHNOLOGIES
[Table of Contents]

At its outset, this study effectively defined an "experiment" as the activity between the time physicists banded together to obtain an experiment number from an accelerator laboratory to the time that physicists stopped trying to publish papers on the basis of what they were allowed to do by virtue of having that experiment number. Embedded in that definition are the assumptions that the criteria for assigning experiment numbers were consistent and corresponded to meaningful changes in experimental research. The post hoc reasonableness of those assumptions is dubious. Among the fixed target experiments in our sample, new experiment numbers were assigned to collaborations that came together to build an experiment "from scratch," collaborations that upgraded some or all apparatus, collaborations that added to their extant apparatus, and to one collaboration that changed virtually nothing and whose members remain perplexed by the assignment of a new number. The collider experiments in our sample were all built from scratch, assigned a number at their inception, and almost never reassigned a number even though they usually investigated several topics and were sometimes upgraded extensively. (The one exception to this generalization was an experiment that received a new number when the detector was moved wholesale, without change, to a different accelerator laboratory; the interviewees considered the data-taking runs on the two accelerators to be all one experiment.)

Given the lack of consistent significance to experiment numbers, experiment-by-experiment comparisons according to the assigned numbers is problematic. However, because fixed-target experiments were usually done in "strings," and because interviewees speak of collaborations that performed strings of experiments as units with social continuity and integrity, it is reasonable to compare how collaborations developed and exploited their experimental capabilities to create strings and to seek an explanation for why physicists performing collider experiments perceived those experiments to be freestanding, that is, independent of predecessors and successors.^[1]

Strings were generated when physicists applied for approval to run a variation in some element(s) of a previous experiment. In our sample of 19 experiments, 15 were fixed-target experiments, and 11 of these 15 can loosely be classified as second or higher in a "string" in the sense that some of the principal organizers of the experiment had worked together on a similar, previous experiment. (Three of these 11 would not qualify as part of a string if the definition were tightened by requiring organizers from more than one institution to have worked together on a previous experiment.) Of the other four fixed-target experiments, two were organized with the intention of becoming foundations for a string and one gave rise to ambitions for starting a string. Only one of the 15 fixed-target experiments in our sample neither stemmed directly from nor directly inspired hopes for another experiment.

The forms of continuity that attracted interviewees' attention were in instrumentation and design. When some of the same physicists received a new experiment number for data runs that reused some or all of the apparatus of a previous run or that used entirely new apparatus in a design that recapitulated or embellished the design of a previous run, those physicists viewed the two experiments as a string. The most common strategy for generating strings of experiments in our sample (11 of 14) was to hold constant the beam and target while varying the detector. Beam- and target-varying strings appear to have been rarer for both administrative and intellectual reasons. Administratively, beamlines were within the jurisdiction of the laboratory and managed in the interests of multiple users, while targets, even though they served individual experiments, had to conform to the safety and design parameters of the accelerator. Thus producing a series of variations in beam or target enmeshed experimenters in additional layers of laboratory management and oversight. Intellectually, beam- or target-varying strings required experimenters to deal with a wider range of phenomena, and their proponents had to be willing to become conversant in a larger number of specialized domains than proponents of strings that varied detectors. In two of the three beam- or target-varying strings, the experiments shifted between measurements of sub-nuclear and nuclear parameters.

Detector-varying strings are divisible by how experimenters modified their detectors and how they justified the modifications. Detectors were modified by either adding components so as to extend the range of data the collaboration could collect or by upgrading extant components so as to increase the efficiency or resolution of the detector. Experimenters justified receiving additional beamtime for these variations by arguing that they would either permit a better investigation of the previous experiment's physics or permit the examination of subjects that the previous experiment had not (or could not possibly have) planned for. There is no correlation apparent between type of detector development and type of justification.

Collaborations performing strings tended to be risk-averse in modifying their detectors. Only two collaborations pursued mid-string detector changes that involved new and risky techniques, and both of these collaborations were adding components to a working core. Collaborations that upgraded individual components or entire detectors only pushed at the limits of the capabilities of current techniques or canvassed manufacturers to find the expertise needed to employ a particularly desirable material. Even the two collaborations that added new and risky components to their detectors had either redundant detection capabilities elsewhere in the detector or a back-up version of the component based on conventional technique. Apparently, collaborations that felt that their working apparatus gave them a solid claim to ongoing beamtime would not risk their credibility with the laboratory administration on making untried technology work, but they could bear the risks of innovation when the new components duplicated conventional apparatus.

Interviewees felt that justifying successive experiments on the basis of new topics to be addressed was the better strategy. One interviewee, who did receive permission to augment his detector in order to examine the same topic as a previous experiment, opined that proposals for such follow-up experiments promise more than they can realistically produce in order to give the follow-up proposal the impact of its predecessor. Another interviewee from a collaboration that had recently rebuilt its detector admired the shrewdness of his collaboration's spokesperson, who presented the case for further beamtime as though the collaboration were on an orderly march through a series of measurements. The interviewee thought that the point of further running was for the collaboration to build up the size of its data sample in order to let the experimenters search for subtle effects once the data were taken. However, the spokesperson recognized that "somehow you can't really say that [to a laboratory's Physics Advisory Committee]. So you say you want to measure this quantity. So he chose a very difficult quantity to measure, which required a lot of statistics." Thus the collaboration got beamtime to improve on its earlier work under the guise of pursuing a particular measurement. (At the end of the run, the collaboration still did not have a large enough sample to make the advertised measurement, "but we did a lot of other things.")^[2]

Whatever their experimental tactics and strategies, the memberships of collaborations performing strings changed greatly over time. However, at the core of every such collaboration in our sample was a small, stable partnership of physicists who dedicated a substantial portion of their research careers to the study and use of particular particles, processes, or techniques. Rather than always starting from scratch to read the direction and data needs of physical theory or to develop detector technologies that could best exploit the next advance in accelerators, these physicists sought improvements to and extensions of their existing experiments, and opportunistically exploited intersections between their subjects and the conceptual framework of contemporary physical theory. Such partnerships of physicists accumulated the experience and resources that attracted others who were "shopping" for an experiment or who discovered they had their own, usually shorter-term interests for doing an experiment in that particular area. The more enlightened and self-secure of these partners granted leadership opportunities to more junior people who had the inspiration and ambition to organize and run an experiment within their bailiwick.

Thus strings appear to be the product of a combination of social, technical, and scientific conditions. When a small set of physicists can form a mutually beneficial partnership to exploit the possibilities of technical flexibility and incremental improvements in the study and use of particular particles or processes, the result is a string of experiments. The fact that nearly all the fixed-target experiments in our sample were parts of strings would seem to indicate a cultural preference among experimentalists for working in this vein. The one fixed-target experiment in our sample that neither stemmed from a preceding experiment nor inspired hopes for a further experiment was socially unique in that it was performed by a collaboration in which theorists and accelerator physicists together outnumbered experimentalists.

Physicists doing experiments at colliding-beam accelerators faced strongly different technical conditions than physicists at fixed-target accelerators. Whereas fixed-target accelerators split and focus their beams on a variety of targets either for direct experimentation or to generate a variety of customized, secondary beams with which physicists could perform experiments, colliders only generated interactions between the one or two particles they accelerate. (All the collider experiments in our sample were performed on electron-positron colliders; henceforth our use of the word "collider" should be understood to refer to electron-positron colliders.) Furthermore, whereas experimenters at fixed-target accelerators stretched their apparatus out linearly behind the target, experimenters at colliders surrounded collision points with concentrically nested detector components because the experimenters' frame of reference and the center-of-mass frame of reference coincided. Consequently, experimenters at colliders were less able to distinguish their experiments by controlling what entered their detectors, and they faced more severe engineering problems and constraints in modifying their apparatus. The continuity typical of fixed-target string collaborations—successive modifications of apparatus in order to delineate all facets of a particle's properties and experimental uses—was technically more difficult to achieve for a collaboration working on a collider experiment. None of the four collider collaborations in our sample pursued a string strategy.

All experiments running simultaneously on a given collider were limited to studying the output of the same collisions, so the four collider collaborations in our sample established their identities by stressing one or another form of detection technique—e.g., tracking of charged particles, detection of neutral particles, or measurement of particle energies—rather than one or another particle or process to be studied. With direct competition to make discoveries, or at least the most precise, meaningful measurements from the same collisions, three of the four collider collaborations tried to maximize further their chances of making a distinctive contribution by developing new technology. One collaboration developed a novel detector component; one used a known technique in a novel design and on an unprecedented scale that together required innovative, difficult assembly procedures; and one used conventional components but in a novel configuration. The fourth sought competitive advantage by shunning novelty in favor of being ready to take meaningful data as soon after the collider turned on as possible.

The unity created in collider collaborations by commitment to developing a detection technique, however, was often limited by differences in the physics interests of collaborators. All the collider collaborations chose or were required to cover as much of the solid angle around the collision point as possible. One collaboration was divided between those interested in events that were primarily detected at low-transverse angles to the collision and those interested in high-transverse events; another collaboration split over the reliability of detection at low-transverse angles, where it was difficult to separate signals from interactions of particles in the two beams from the passage of non-interacting particles in the beams. And all collider collaborations had to reach a position on what energy the accelerator should be set for upcoming runs. Because electron-positron annihilations create a state of pure energy equal to the sum of the energies of the colliding particles, there is no possibility of investigating particles (and their decays) that are not created at the energy of the collisions. Two collaborations were divided over the best setting for the accelerator because collaborators wished to use the detector to study different particles.

Collider collaborations did not perform strings because they could not readily exercise discrimination over what entered their detectors, nor readily add to or modify the components in their detector, and thus their principals could not unite around examining in different ways the physics and experimental uses of particular particles. Instead, collider collaborations were coalitions held together by a common interest in a particular detection technique and a particular accelerator. The demise of either spelled the end of the collaboration. While none of the collaborations in our sample experienced detector failure, accelerator laboratories have let older accelerators deteriorate as they shifted their best personnel to the creation of a new accelerator. Interviewees in all four collaborations pointed to either the decline of the accelerator or their inability to control its running energies as ending their experiments before data-taking possibilities had been exhausted.

V. ROLE OF SPOKESPERSON
[Table of Contents]

Narrowly speaking, a spokesperson has been an administrative convenience, an individual designated to speak for the collaboration to laboratory management and to inform collaborators of laboratory requirements. However, the first communication between collaboration and laboratory management—the presentation of the experiment proposal to the laboratory's Physics Advisory Committee—has been appropriate for a collaboration's intellectual leader, the physicist who had thought most thoroughly about the experiment and invested him/herself most deeply in seeing the experiment performed. Subsequent communication rarely required the conveyance of intellectual passions and strategies, but in the words of one interviewee, "it's certainly a hell of a lot better if the spokesperson does offer intellectual leadership, because if you don't know what you're talking about to a lab, you really can't get the lab to understand what you want." Thus collaborations usually made an experiment's instigator their spokesperson. The role of spokesperson, another practitioner observed, was both cherished as a symbol of scientific initiative and despised for the administrative tasks that come with it. While not all physicists disliked serving as administrators, collaborations in the later period covered in this study appear to have generated more managerial tasks, and the meaning of being spokesperson and the role the spokesperson plays seem to be in flux.

In the 14 fixed-target experiments in our sample, the intra-collaboration role for the spokesperson after the experiment's approval was generally to coordinate the activities of the collaborators and to oversee the installation of the apparatus at the laboratory. These experiments did not generate issues that made the spokesperson's duties seem particularly critical to the development of the experiment, in his/her own or collaborators' eyes. The spokesperson for one of these experiments said nothing in his interview about being spokesperson, and a non-spokesperson interviewed for a different experiment could not recall who the spokesperson was. Others spoke of "collective decision-making" or "pure democracy" as the practiced form of governance. When hierarchies based on professional rank were evident to the participants, the spokesperson "had no particular role" beyond that of other group leaders. When the office of spokesperson for a collaboration did not function well, as was the case in the collaboration where the two co-spokespersons would reverse each other's decisions in their competition to stamp their personalities on the experiment, the collaborators still managed to build a detector that out-performed its competitor at the laboratory.

While the spokesperson's role in fixed-target experiments, even when less than efficiently filled, did not make or break any of the experiments in our sample, only once did any spokesperson of a fixed-target experiment relinquish the role as more trouble than it was worth. Spokespersons did not retain their title because the duties were light; several commented on the burden of familiarizing themselves with all aspects of what they were coordinating. Rather the task of coordinating collaborators was itself evidence of scientific leadership and initiative. Because collaborations had so few powers to reward and discipline their members, spokespersons had to reason and persuade their way through the conflicts and misunderstandings that inevitably arose, and retention of their position was evidence of their skills of persuasion. Junior faculty on the experiments in our sample thus particularly coveted the office of spokesperson in the belief that it would help their tenure campaigns; five of the collaborations had junior-faculty spokespersons for the experiments in our sample, and a sixth had a junior faculty spokesperson for a later experiment in a string.

The four collider experiments and three of the four fixed-target experiments approved in or after 1979 and not part of a string according to the more restrictive definition (see p. 14) contained managerial practices not found in the earlier fixed-target experiments. All the collider experiments and two of the three fixed-target experiments had some sort of administrative substructure, such as deputy spokespersons, executive committees, or ad hoc committees, to handle collaboration business. Three of the four collider experiments shifted spokespersons over the course of their runs—one on an annual basis—while the third fixed-target experiment changed spokespersons because the initial spokesperson, chosen for his administrative abilities and favorable institutional position, resigned as spokesperson when he changed jobs. All these collaborations apparently found the administrative burdens of overseeing an experiment to require social innovation.

The management of collaborations has certainly needed to be strengthened as the drive to produce more refined, sophisticated measurements required larger, more complex detectors that absorb the design, construction, and analysis efforts of more physicists. In most of the more recent experiments, neither the individual spokesperson nor the collaboration as a whole could coordinate and evaluate the many tasks of the collaboration's members. Thus they resorted to the use of committees or additional officers to handle some collaboration business. And thus even experiments that did not create substructures showed signs of needing them, as when an autocratic spokesperson delegated authority to a junior colleague, or when a loosely run collaboration lost track of the status of its individual members' projects.

The practices of shifting spokespersons and selecting them for their administrative skills or position also suggest that the characteristics of some experiments and collaborations required that managerial skills be placed ahead of intellectual initiative in collaboration governance. This seems particularly true for collider experiments, whose geometry of concentrically nested components made them more tricky to build and whose acceptance of all events emanating from collisions led to more possibilities for experiments within experiments. In one collider experiment, the initial spokesperson, whose invention of a new kind of detector component was the collaborations's raison d'être, resigned because he found himself to be a poor technical coordinator and wanted to concentrate on the problems of the particular component he had invented. In another, senior collaborators at the outset recruited an engineer to manage the coordination and assembly of components. In a third, the spokesperson retained his position, which he described as chief-executive-officer for construction, but relied heavily on two senior collaborators for help in dealing with a unique industrial supplier whose product was not meeting specifications. And all four collaborations encountered managerial problems because their members were divided by their interests in various types of physics or their willingness to trust components that detected particles traveling parallel and close to the beam. When an experiment consisted of an intricate construction project followed by the investigation of several intellectually distinct physics topics, there was little role for enduring leadership, and shifting the burden of managing such stressful operations and internally divided organizations may keep resentments towards any individual from accumulating.

VI. ORGANIZATION FOR DETECTOR CONSTRUCTION, DATA ANALYSIS, AND COMPUTER PROGRAMMING
[Table of Contents]

The primary factors in producing a collaboration's organization for building a detector were logistical convenience, the availability of appropriate personnel, and technical experience. Large components and delicate components were best built near or at the accelerator to minimize transport problems. Tricky components were best designed and built at institutions with high-quality engineering staffs, while mass-produced, labor-intensive components were best built at institutions with access to inexpensive, often undergraduate help. Also when particular physicists had prior success in building a similar type of component, they tended to recapitulate their earlier success in their later experiment. Rarely did interviewees speak of being "end-use driven," that is, building particular components because they expected them to be strategic in examining a particular branch of physics. And rarely did collaboration organizers report that they consciously evaluated possible individual collaborators with an eye to acquiring a balanced set of physics interests that was suitable to building a particular detector.

Although responsibilities for building parts of the detector were often assigned or claimed independently of physics interests, these responsibilities nevertheless had medium-term ramifications in some collaborations. A large task for many of the collaborations in our sample was the reconstitution of the mass of digitized, electrical pulses, which made up the experiments' raw data, into categories and events that physicists could use to make calculations or measurements. The group that built a component would typically be responsible for writing the software that performed this "pre-analysis," and the people who did this work would find they had an initial comparative advantage over others in working on physics topics that made heavy use of their component. Such was most often the case in multitopic experiments on which students from more than one university were simultaneously writing dissertations. However, when experiments ran for longer than was needed to collect data for "one generation" of graduate students and postdocs, comparative advantages based on detector-construction assignments tended to dissipate. The new students and postdocs would be initiated into the experiment in a less specialized fashion, and they would want to pursue more complex topics that combined the demonstrated capabilities of the several detector components.

The quantity of work that went into writing the one-of-a-kind programs to convert the signals from individual detector components into physics information insured that collaborations building a many-component detector for a multi-purpose experiment would treat those programs as communal property that all should use and nobody should duplicate. However, for the more tightly focussed electronic experiments and the experiments that collected data on film, the possibility of multiple analyses of raw data was real. Furthermore, with the advent of interactive data-analysis programs that permitted physicists to process data from their terminals rather than submit batch jobs, even physicists in many-component experiments efficiently performed "cuts" on the data to create individualized data samples at a higher level of generality than previously. These conditions posed a "federalist" dilemma to collaborations: to what extent should reconstruction programs and general data samples be subject to central collaboration authority to insure homogeneity and commensurability of results reached by physicists from different institutions; and to what extent should local customs in data management be allowed to flourish in order to cross-check results and to support the widest possible spectrum of tastes and interests? The former policy, by being more efficient, would seem preferable when a collaboration is in heated competition with others to reach an exciting result, while the latter, by being less prone to allow collaboration-wide errors, would seem preferable when a collaboration has a physics niche more to itself. However, the practices of individual collaborations appear to have hinged more on the personal preferences and working relations among a collaboration's leaders than on the external conditions collaborations faced. Some physicists saw anarchy in the use of multiple programs at a general level and trusted in their collaborators' abilities to spot any flaws in widely used programs; others saw dictation in the collaboration-wide use of someone's favored set of general programs and trusted in their collaborators' abilities to argue their way professionally to a consensus should the use of multiple programs yield conflicting results.

Because computer programming has been so entwined with the extraction of physics from the data, experimentalists have largely done their own programming and have not relied heavily on trained specialists in computer science or engineering. Some interviewees viewed themselves as "computer-oriented" and specialized either in the data acquisition problems of getting all the electrical pulses onto identifiable parts of magnetic tape or in the software-management problem of making sure the programs to handle the data from the several detector components work together harmoniously. Our sample contains fragmentary evidence that computer-oriented physics has become a viable, informal sub-specialty analogous to specialization in the design and construction of certain kinds of detector components. Interviewees from two of the collaborations in our sample spoke appreciatively of innovations in computation that were made by individuals who had made a point of thinking about the common factors in experimenters' uses of computers rather than just the needs of a particular experiment. However, collaboration organizers did not generally have to make a point of finding people who already had or were willing to acquire the computation skills an experiment needed.

VII. INTERNATIONAL COLLABORATIONS
[Table of Contents]

International collaborations appear to be children of necessity. None of the international collaborations in our sample originated in prior personal or professional contacts as was commonly the case with domestic collaborations. From the perspective of U.S. experimentalists (which is the perspective of most of our interviewees) any of four factors behooved experimentalists to seek foreign collaborators. First, a foreign group had developed an experimental technique that U.S. physicists wished to use and learn. Second, an experiment required more manpower and money than could be readily raised domestically. Third, a laboratory director spotted common interests in proposals from domestic and foreign collaborations and brokered a merger of the two. And fourth, U.S. experimenters with a working detector desired more beamtime than a U.S. accelerator had the will or ability to provide. For the non-American experimentalists, collaborating with Americans was a means for performing experiments that the foreigners' domestic infrastructure could not support.

Four kinds of problems, beyond the obvious ones of language, appear in international collaborations: technical, cultural, logistical, and political-legal. The experimenters were left to their own devices to deal with technical and cultural problems; logistical and political-legal issues involved people outside the collaboration.

Technical problems cropped up in collaborations that had their origins in the complementary expertise and resources of the participating institutions. In these collaborations, the foreign physicists preferred to build their components in their own laboratories and shops. When the foreign physicists' nation employed different metrics or standards from the U.S.A., the integration of the components with the rest of the detector and the American accelerator required additional work. That work was well within the power of physicists to perform, and there were no examples of such problems not being routinely dispatched.

Cultural tensions were most noticeable in U.S.-Japan collaborations. This condition may be due as much to the recentness of Japan's large-scale entry into high-energy physics as to the depth of cultural differences between Americans and Japanese. Both of the U.S.-Japan collaborations in our sample resulted from over-extended U.S. collaborations with approved experiments using the 1979 Implementing Arrangement on the US/JAPAN Cooperation in the Field of High Energy Physics to acquire additional money and people. In neither case had there been prior institutional ties among these American and Japanese groups; nor did the process of pulling together a proposal provide an opportunity for any of the physicists to grow accustomed to dealing with each other before the pressures of building and assembling components threw them into intense and close working relationships.

Japanese and American physicists from both of the U.S.-Japan collaborations in our sample noted that Americans argued more vociferously and showed less deference to hierarchical relationships and that the Japanese held more of their discussions outside formal meetings. In one collaboration, detector designs were so far along before the Japanese became involved that the Japanese physicists joined American groups and did not build anything as a group unto themselves, while in the other, the Japanese did build a particular piece of the detector as a group. However, it is difficult to generalize on the consequences of these arrangements for easing cultural tensions within the collaborations. In the former collaboration, a U.S. physicist still felt disoriented by Japanese processes for reaching consensus or differentiating open-ended projects from delimited tasks even though the Japanese were not operating as a group. In the latter collaboration, a Japanese physicist still felt burdened at having to be more vocal than was comfortable for him even though he was part of a Japanese team. The significance of these differing arrangements may reside more in their subsequent consequences for Japanese physicists than in their impact on how these collaborations performed their experiments. Two American physicists from the collaboration where the Japanese did not function as a group expressed concern that the Japanese physicists became "over-Americanized" and were having trouble doing science through Japanese institutions, but the senior Japanese physicist in this collaboration was glad that his juniors had to work so closely with Americans because he wants high-energy physics to be part of a general opening of Japanese society to international influences.^[3]

Cultural differences, though less pronounced and less common, did sometimes surface between West European and American physicists. The differences lay not in the group dynamics of collaboration meetings but in contrasting assumptions over the degree to which collaboration-wide rules, especially for handling raw data and reconstructing events, should be instituted over the autonomy of the participating institutions. One European-American collaboration, which was formed at the behest of the accelerator laboratory's director, spent two years arguing over whether a pan-European system for reconstructing events in this kind of experiment should be the norm for the collaboration or whether all participating institutions should retain the privilege of using and modifying their own programs. The Europeans apparently won by strength of will borne of having already forsworn local autonomy in the course of creating the pan-European system. In another all-American collaboration, a European who was working for one of the American institutions feared the dispersed physicists would reach incompatible, incommensurable results because the collaboration was lax in instituting collaboration-wide controls over basic analysis programs; he did his best to make the on-line programming a structure for post-run analysis.^[4]

The logistical problems of an international collaboration usually obliged the foreign groups to pick up and move to the experiment's site. That strategy could have posed severe fiscal and administrative problems stemming from exchange rates and the rules that governments impose on the use of public monies, but U.S. laboratory administrators were flexible about juggling laboratory funds in order to cover those things that foreign physicists needed but could not buy and not to cover those things that foreign physicists provided for the whole experiment. Only one of the international collaborations in our sample was successful in maintaining significant work at several sites and using meetings and communiqués to keep everyone informed. Two others tried to have at least multiple centers for data analysis: one gave up because the foreigners could not readily keep the analysis software up to date, while the other persisted but ended up publicizing erroneous results because one group could not readily scrutinize the work of the other. One set of collaboration organizers, when confronted with the interest of a foreign physicist who had independently designed an experiment similar to theirs, agreed only to accept his participation as an individual and rejected having his university as an institutional collaborator because "it would have just made our life unnecessarily more complicated."

The potential for political-legal problems has been most spectacular in American-Soviet collaborations, (though one West European physicist required a laboratory director's help with immigration authorities when the experiment lasted longer than his visa). The physicists' primary fear was that their governments would treat them as pawns in diplomatic maneuvers over events outside of physics, and their primary response was to "conspire together to fight each other's [governmental] bureaucracies." But more importantly, the collaborations' leaders built physics-based friendships that could survive the diplomatic problems that physicists themselves created; two East-Bloc physicists from experiments in our sample defected—one after and one during their experiments—but the leaders of their groups kept up their relationships with American physicists. Less spectacular, politically-based problems stem from differences in science policies of the nations of collaborating physicists. One international collaboration encountered difficulties in its formation because the experiment was slated for running on an accelerator in the process of being upgraded; while the high-energy physics community of the nation that was investing in the improved accelerator could not politically afford to let technical uncertainties about the upgrade inhibit the planning of experiments, the physicists of other nations had a much more formidable job convincing their administrators of the wisdom of committing resources to an experiment on an unproven accelerator. In another international collaboration, physicists of different nations had to contend with the sense that they needed different types of accomplishments to maximize their chances of winning governmental support for future experiments.

Despite the difficulties of working in international collaborations, almost all the participants found that the value of building relationships with foreign physicists more than outweighed the difficulties. Time and experience should erode cultural misunderstandings, and further innovations in communication and transportation may alleviate logistical burdens. But as experiments grow larger, longer, and more expensive, legal and political impediments to participating in a foreign experiment can only grow more threatening. Not only will physicists need visas and travel privileges, but also their spouses will probably want the right to pursue their careers during a long stay in a foreign country. Not only will physicists have to argue that an experiment is likely to make a significant contribution to physics, but also that their participation in an experiment will fit into a broader framework of policy for the use of national resources for research.

VIII. DISSEMINATION OF RESULTS
[Table of Contents]

Interviewees for six experiments from the early period of this study made little or no mention of conference talks. For the rest, conference talks were an important, though not always rare, commodity that a collaboration exercised care in distributing. Talks were invariably considered valuable for bringing results quickly to the wider community's attention, though there was variance on the standards of review collaborations imposed on talks as compared to publications. And talks were invariably considered valuable for conferring credit and granting exposure to the collaboration's lesser-known members who had driven forward a particular line of analysis—though just who decided the deserving individual varied across collaborations.

Although the question set was not designed to elicit the standards that collaborations set for conference talks, some variations in opinion are apparent. Two collaborations imposed less stringent standards on talks than on publications. In one, "a result that you think is right within reason" could be presented at a conference while journal articles had to be as trustworthy as possible for future physicists. In the other, laboratory seminars were used to try out questionable results or ideas. By contrast, interviewees from three experiments reported that the collaborations rigorously reviewed the content of at least some talks; in one case, a collaboration waited several months to present a central result to the accelerator laboratory's seminar, even though the collaborators felt certain about the result, while they worked at making the result as convincing as possible. Such discipline, however, did prove difficult to enforce pleasantly. One of the collaborations suffered from "bad citizens giving impromptu talks" while in another "a bit of feeling" was aroused when one collaborator blocked another from giving a seminar at an accelerator lab.

Except when choosing someone to present important results at major conferences—in which case the presenter's reputation and speaking abilities as well as his contribution to the results were taken into account—most collaborations did not use collaboration meetings to decide who spoke at conferences. A variety of ways were used for delegating such decisions. In two collaborations, the spokesperson either openly or effectively decided; more commonly, the group leaders collectively made the decision; when institutional lines remained meaningful through the course of an experiment, conference presentations were sometimes granted to institutions with each group responsible for deciding who would speak. Two collaborations did collectively decide on conference speakers. In one, the responsibility for analyses was well enough defined that there was rarely any issue over who should present results; in the other, most collaborators were concerned with one central topic, and collaboration-wide discussion insured equity and curbed "a few individuals who like to give all the talks." In general, where disputes over who gave presentations was occasional, it seems collaborations delegated responsibility for making those decisions, but where such disputes were rare or frequent, collaborations collectively made those decisions.

For the production of journal articles, the power of writing the first draft was inconsequential in the overwhelming majority of experiments. In only one instance, where the experiment's leaders felt themselves to be in a race to claim a discovery, has anyone reported that the mass of collaboration members were denied the opportunity to criticize an initial draft. For the experiments approved before 1976, there were no formal intermediaries between a paper's drafter and the collaboration as a whole. But in six of eleven experiments approved since 1976, collaborations have created committees to work with paper drafters before or after the collaboration as a whole has discussed the results. As collaborations have grown larger and more geographically far-flung, there are obvious economies in minimizing the number of times a collaboration meets to discuss the merits of a particular proposed publication, but there are no neat criteria for distinguishing the collaborations that did and did not institute such procedures.

Interviewees reported instances where papers sailed through intra-collaboration review with one set of revisions and instances where the collaboration insisted that the initial drafters start again from scratch. The general ethos, which has remained constant over time, is that consensus within the collaboration should be reached before a paper is submitted to a journal with a paper's detractors subject to the constraint that "you can't be the only one, and if you are, you better not be it each time." There were no reports of individuals in the experiments under study asking to be removed from author lists out of distrust of a paper's results; indeed, one interviewee recalled that the collaboration averaged the results of differing, independently-produced analyses rather than either putting off publication or publishing separate papers. (There are reports of people taking their names off papers from other experiments.) In experiments that effectively contained experiments within experiments, collaborators have removed their names from author lists from a sense of not having contributed to the results; such actions may indicate a belief that the work was not worth the effort to produce it.

At least five of the collaborations examined have tried to recognize an individual's contributions to a particular analysis by placing him at the head of the author list. Discussions of who belonged in that position, to an even greater extent than who deserves to present a paper at a conference, generated "more heat than light," and one veteran of such discussions recommended that they be scheduled for just before lunch or dinner, when everyone is eager to adjourn. One collaboration that stuck to recognizing individuals with first place on an author list decided that the results of two independently-written doctoral theses should be included in one article; the ensuing debate over who should be listed first was "the most acrimonious" in a collaboration notable for a high degree of internal competition and fierce debates over data interpretation. Those collaborations that used an alphabetized author list did not have such headaches, and one spokesperson actually preferred alphabetized lists because they drove home the point that "there's more to doing an experiment than doing a particular analysis." Besides occasionally stressing first place on an author list, collaborations sometimes tried to boost the significance of the author list by holding down the list's length through the exclusion of engineers or through rules that set how long a physicist could remain an author after he left the collaboration. (Such rules also protected physicists who left an experiment before significant data came in from losing credit for their efforts to design and build the experiment.) However, the overwhelming impression given by the interviews is that reputations in high energy physics have been built through word-of-mouth, letter of recommendation, and participation in conferences.

For two-thirds of those collaborations whose members expressed an opinion, Physical Review Letters was and remains the journal of choice. But a minority preferred or has come to prefer Physics Letters, because it turns around manuscripts more quickly and is more flexible on issues of article length.

IX. ORGANIZATIONAL STRATEGIES AND COMMUNICATION
[Table of Contents]

Basic institutional and intellectual conditions of high-energy physics experiments required that all collaborations combine three organizational strategies, each with a characteristic pattern of communication. First, because detectors must be assembled and data taken at a laboratory, the laboratory had to be treated as an organizational headquarters to which the outlying institutions passed and received information. Second, because no single institution had the resources to mount an experiment on its own, and because reproducibility of results was an essential confidence-builder when puzzling or controversial findings were claimed, labor had to be divided and duplicated with collaborators working independently and reporting to the collaboration as a whole their progress, methods, and results. Third, because any individual's research could make use of equipment he had not built and software he had not written, collaborators had to contribute to an information pool that enabled each to take full advantage of what others had developed.

Individually, collaborations faced a host of technical, institutional, historical, and geographic factors that made for noteworthy variation in the difficulties they found most problematic. Each collaboration idiosyncratically blended complementary organizational strategies and compromised among conflicting strategies in order to have the best chance of handling its toughest difficulties. However, no collaboration could (happily) allow its toughest difficulties to drive the shaping of strategy to the exclusion of accommodating the interests of those with lesser difficulties. Technical factors were an obvious source of compelling problems for collaborations, but the other factors also significantly influenced the shaping of strategy—sometimes in ways that ran counter to technical factors.

Viewed over time, our sample brings out one major trend and two major continuities. The creation of intra-collaboration information was increasingly formal (e.g., collaboration-wide mailings and memoranda) and increasingly electronic in the larger, more recent experiments. Collaborations continued to divide labor, and the collaboration meeting remained the forum for discussions that led to decisions concerning the tactics and results of experiments. Even interviewees who found collaboration meetings unpleasant did not suggest alternatives to their use to debate and decide the physics issues in an experiment.

A. Technical Factors
[Table of Contents]

When detector components fit snugly together, as was commonly the case with collider detectors, the need to insure adequate space for the components and their auxiliary equipment (e.g. mountings and cables), and adequate access to components that require maintenance, often drove collaborations to establish a headquarters at the accelerator laboratory. Collaborators stationed there, whether laboratory staff or university employees, could thus continuously monitor the construction and assembly of detector components. In one collider experiment, the laboratory-based spokesperson viewed himself as the chief-executive-officer for construction and tried to give collaborators away from the laboratory a feeling of involvement by writing and distributing minutes for meetings. (His efforts were not entirely successful in the eyes of one student, who felt out of touch with the experiment as a whole while working away from the laboratory.) In two other collider experiments, collaborators who undertook significant construction away from the accelerator spoke appreciatively of an engineer at the accelerator laboratory who tracked the progress of the several components and made sure that incremental changes in the components' designs did not make them impossible to assemble or maintain. In others, physicists simply noted the necessity of living at or frequently commuting to the experiment site in order to contribute to the material creation of an experiment.

By contrast, when the detector was unimposing and the greatest technical challenge to the experiment lay in the development of target or beamline, there was usually much inter-institutional communication between the accelerator staff and external proponents of the experiment. In one such experiment, regularly scheduled telexes between the accelerator laboratory's research group and the target builders kept both institutions technically coordinated. In another, collaboration meetings were needed only sporadically because of the volume of telephoning among the participating physicists.

Finally, when detector components individually presented challenges, but problems with their integration seemed manageable through on-the-spot adjustments, a division of labor was instituted. In these experiments, regular meetings were the rule and frequent memorandum writing was encouraged so that all collaborators became sufficiently well versed in the components they were not building to understand the limitations in the data they produced.

Regardless of how communications were organized during construction, data analysis was usually handled through a headquarters framework. In many experiments, the collaboration possessed little discretion; one group (most often an accelerator laboratory research group) had easier access to the computer facilities needed to handle the quantity of collected data, and that group's institution became the headquarters for data analysis. Even when a collaboration included more than one institution with adequate computer facilities, one institution still often became data-analysis headquarters in order to insure most easily that all data were being analyzed with the most up-to-date programs. In the case of bubble

AIP STUDY OF MULTI-INSTITUTIONAL COLLABORATIONS
PHASE I: HIGH-ENERGY PHYSICS

REPORT NO. 4:
HISTORICAL FINDINGS ON COLLABORATIONS IN HIGH-ENERGY PHYSICS

By
Joel Genuth, Peter Galison, John Krige, Frederick Nebeker, Lynn Maloney

TABLE OF CONTENTS

AIP STUDY OF MULTI-INSTITUTIONAL COLLABORATIONS PHASE I: HIGH-ENERGY PHYSICS

REPORT NO. 4: HISTORICAL FINDINGS ON COLLABORATIONS IN HIGH-ENERGY PHYSICS

By Joel Genuth, Peter Galison, John Krige, Frederick Nebeker, Lynn Maloney

TABLE OF CONTENTS

AIP STUDY OF MULTI-INSTITUTIONAL COLLABORATIONS
PHASE I: HIGH-ENERGY PHYSICS

REPORT NO. 4:
HISTORICAL FINDINGS ON COLLABORATIONS IN HIGH-ENERGY PHYSICS

By
Joel Genuth, Peter Galison, John Krige, Frederick Nebeker, Lynn Maloney