Instrument capture is commonly attributed to scientific pathology — groupthink, intellectual laziness, paradigm entrenchment. This paper locates its cause elsewhere: in the rational response of individual scientists to the actual incentive structure they inhabit. Publication metrics reward positive results, established instruments, and rapid output. Grant structures fund paradigm-aligned proposals from established research groups. Career advancement selects for publications over discoveries. Within this architecture, instrument capture is not a failure. It is the correct strategy. Three domain studies — CERN-scale physics, clinical psychiatry, and consciousness research — document the specific financial and career logic through which each field arrived at capture as its equilibrium state.
The Structural Question
The standard explanation for instrument capture runs through the sociology of knowledge: scientific communities develop investment in existing tools, training pipelines entrench measurement paradigms, and the cost of abandoning sunk infrastructure — both material and conceptual — is too high to absorb even when anomalies accumulate. This is the Kuhnian frame (1962). It is not wrong. But it does not account for why individual scientists, who are not required to be paradigm defenders, systematically choose paradigm-aligned research strategies even when they have private knowledge that the dominant instruments are limiting.
The answer lies not in sociology but in economics. The funding and publication systems that govern scientific careers create a specific incentive architecture — a set of reward structures and career consequences — that makes instrument capture the rational choice for any scientist who wishes to maintain a productive career. Understanding this architecture is not an indictment of scientists as individuals. It is a diagnosis of the system they correctly navigate.
Instrument capture is not primarily a failure of scientific culture, integrity, or epistemology. It is the predictable output of an incentive architecture that rewards the use of established instruments over the development of better ones, positive results over null results, and incremental publications over paradigm-challenging questions. Within this architecture, capture is not irrational. It is optimal.
The structural question this paper addresses is therefore not "why do scientists get captured?" but "what would the incentive architecture have to look like for scientists not to get captured?" The answer to that question reveals the distance between the current system and any system that could support instrument independence as a stable equilibrium.
Instrument Capture Defined
Instrument capture, as developed in earlier papers in this series, refers to the process by which a scientific field inverts its relationship to its measurement tools. The normal relationship: instruments are designed to answer questions. The captured relationship: questions are designed to fit instruments. The field continues to produce output — publications, findings, citations — but the output is constrained to what the existing instrument can register. Questions that require different instruments are excluded from the researchable domain, not by explicit prohibition but by the practical consequences of pursuing them.
The captured state is self-concealing. Within a captured field, the instrument appears comprehensive because all findings are generated through it. The questions it cannot answer are invisible, because they are never asked with sufficient institutional resources to produce publishable results. The field looks productive. The incompleteness of the instrument is not in the data; it is in the shape of the data that isn't there.
The research question determines what instruments are needed. Instruments are chosen, adapted, or invented to answer the question. The question governs.
The available instrument determines what questions can be asked. Questions are chosen to fit existing measurement capacity. The instrument governs. The field optimizes within the constraint.
The mechanism by which capture is maintained is the subject of this paper. The Instrument Capture Loop (ICS-2026-EI-002) documented the epistemological dynamics. This paper documents the financial and career dynamics: the specific reward structures that make the captured equilibrium stable against internal dissent.
The Publication Architecture
Scientific career advancement is governed, in every major research system, by publication output. The currency is not discoveries but papers: peer-reviewed publications in journals with recognized impact factors. This creates a specific incentive structure that shapes every decision a working scientist makes about what to study, how to study it, and what to do with the results.
The publication bias problem is the most documented pathology. Journals systematically prefer positive results — findings that confirm or extend existing knowledge — over null results, replications, and anomalies. Fanelli (2012) analyzed 4,656 papers from 1990 to 2007 and found that the proportion of papers reporting positive results increased significantly over the period, across all fields, at a rate inconsistent with genuine increases in research success. The increase was concentrated in fields with higher susceptibility to flexibility in data analysis and outcome reporting. The publication system was actively selecting for positive outcomes independent of whether positive outcomes were true.
An experiment that confirms the predictions of an established instrument generates a publishable result. An experiment that produces a null result — the instrument failed to detect the expected effect — generates either a rejection slip or a file drawer. An experiment that produces an anomaly requiring a different instrument generates either a methodological controversy or, more likely, a recategorization as measurement error. The publication architecture rewards the first outcome, penalizes the second, and eliminates the third.
The citation economy compounds the effect. Impact factors — the metric used to rank journal prestige — are computed from citation rates. Papers that cite established instruments in established frameworks receive more citations than papers that challenge them, because more researchers are building on the established framework. High-impact publications beget high-impact citations. The scientist who challenges the dominant instrument faces not just publication difficulty but citation isolation: even if the challenge paper is published, it will accumulate fewer citations than confirmatory work in the same field, because fewer researchers are working in the challenger's paradigm. The citation economy naturalizes the dominant instrument's supremacy.
The p-hacking and flexibility problem follows directly. Given that positive results are required for career advancement and publication success, scientists under career pressure have strong incentives to use the analytic flexibility available in most research designs to produce positive results. Simmons, Nelson, and Simonsohn (2011) demonstrated that through legitimate-seeming choices about data collection, exclusion, and analysis, researchers can increase the probability of a false positive to above 60%. This is not fraud — it is the rational use of analytic discretion under publication pressure. The publication architecture does not require scientists to lie. It creates conditions in which the truth and career success are systematically misaligned.
The Funding Architecture
Scientific funding is governed by grant competition. In the United States, federal funding agencies — NIH, NSF, DOE, DoD — account for the majority of academic research funding. Grant awards are determined by peer review: panels of active researchers evaluate proposals and assign scores that determine funding priority. This system has a structural consequence that is rarely acknowledged: peer review panels are composed predominantly of researchers who have built their careers within established paradigms, using established instruments. They evaluate proposals for research that challenges those instruments.
The bias is not intentional. It is structural. A proposal to develop a fundamentally new measurement approach in a field where established instruments dominate will be evaluated by reviewers who (a) have expertise in the established instruments, (b) have career investments in the established paradigm, and (c) lack expertise in the proposed alternative. The proposal will be judged primarily on criteria — methodological rigor, theoretical grounding, preliminary data — all of which favor alignment with the existing paradigm. The preliminary data criterion is especially consequential: novel instruments by definition cannot have extensive preliminary data in established journals. The grant system systematically defunds instrument development before it can demonstrate viability.
To develop a new instrument, you need funding. To get funding, you need preliminary data in peer-reviewed publications. To get peer-reviewed publications, you need a functioning instrument. To get a functioning instrument, you need funding. The catch-22 means that instrument development can only be funded through institutional support independent of the standard peer review process — which is rare, discretionary, and typically available only to researchers with pre-existing institutional capital within the paradigm they are trying to challenge.
The programmatic funding structure adds a second layer. Major federal agencies fund research programs — sustained, multi-year initiatives around specific scientific questions and methods — rather than isolated individual grants. Once a program is established, its funding creates a constituency: researchers trained in the program's methods, institutions that have built infrastructure for the program's instruments, and review panels drawn from the program's alumni. Programmatic funding self-perpetuates. The BRAIN Initiative, the Human Genome Project, the SSC and its successor programs — each creates a funding ecosystem that is structurally difficult to redirect once established, because redirection requires defunding the people most qualified to evaluate the program's progress.
Domain Study: High-Energy Physics and the Infrastructure Lock
High-energy particle physics is the clearest case of infrastructure-driven instrument capture, because the instruments are physically enormous and correspondingly expensive. CERN's Large Hadron Collider — the dominant instrument for particle physics since 2008 — cost approximately €6.4 billion to construct, employs thousands of researchers globally, and requires hundreds of millions of euros annually to operate. It is the most expensive scientific instrument ever built. This creates an incentive architecture of unusual clarity.
The physics questions the LHC can address are bounded by what it can measure: particle interactions at energy scales up to approximately 14 TeV. Physics questions that require different energy scales, different particle types, or different detection modalities are not addressable by the existing instrument. But the LHC's existence and cost generate overwhelming pressure to keep it running and to define the central questions of particle physics as the ones it can answer. The $30B+ in sunk costs across global HEP infrastructure is not merely financial — it represents the career investments of thousands of physicists trained on, employed by, and publishing with these instruments. The scientific community most qualified to evaluate whether the LHC paradigm is limiting is the community most invested in its continuation.
Lee Smolin's analysis in The Trouble with Physics (2006) documents how this dynamic played out in string theory's dominance of theoretical physics: a mathematical framework that makes no testable predictions at accessible energy scales became the dominant career path for theoretical physicists, not because it was empirically productive but because it generated the publication currency — papers, citations, academic positions — that the incentive architecture rewards. String theory became the answer to the question "what do you work on to build a career in theoretical physics?" rather than "what do you work on to understand nature?" The instrument (mathematical formalism) captured the question.
Domain Study: Clinical Psychiatry and the DSM-Pharma Interface
Clinical psychiatry's primary instrument is the Diagnostic and Statistical Manual of Mental Disorders — the DSM. The DSM provides the categorical diagnostic system that defines what conditions exist, how they are identified, and by implication, what interventions are appropriate. It is not merely a measurement tool; it is the ontological infrastructure of clinical psychiatry. Its categories determine what insurance reimburses, what pharmaceutical companies develop drugs for, what clinical trials can enroll, and what research questions generate fundable proposals.
The incentive architecture connecting DSM categories to pharmaceutical investment is unusually explicit. Drug development is organized around DSM diagnostic categories: a compound that reduces scores on a validated DSM-aligned symptom scale in a randomized trial will receive FDA approval and market access. A compound that addresses underlying neurobiology without mapping cleanly to a DSM category will not. Pharmaceutical companies therefore fund research that validates and extends DSM categories, because DSM alignment is the path to regulatory approval. They do not fund research that challenges DSM categories, because category challenge threatens the validity of the diagnostic instrument their products are approved to treat.
The DSM revision process is itself shaped by pharmaceutical incentives in documented ways. Cosgrove et al. (2006) found that 56% of DSM-IV panel members had one or more financial relationships with pharmaceutical companies. In panels covering categories directly treated by approved drugs — mood disorders, psychotic disorders, anxiety disorders — the rate was substantially higher. The instrument that defines psychiatric disease categories was substantially developed by researchers with financial relationships to companies whose products are indexed to those categories. The incentive architecture does not merely shape which questions get asked. It shapes which conditions get defined.
Research Diagnostic Criteria alternatives, the National Institute of Mental Health's Research Domain Criteria (RDoC) initiative, and neurobiological dimensional approaches all represent attempts to develop instruments that escape DSM category dependency. All have faced the same structural obstacle: funding agencies and pharmaceutical companies cannot sponsor clinical trials using non-DSM instruments, because regulatory approval pathways require DSM-aligned outcome measures. The alternative instrument has no pathway to the career rewards — publications, grants, drug development partnerships — that sustain academic psychiatry. The dominant instrument retains its position not because it is better than alternatives but because it is wired into the funding infrastructure that makes scientific careers possible.
Domain Study: Consciousness Research and the Paradigm Dependency
Consciousness research represents a different variant of the incentive problem: a field where the dominant paradigm (neural correlates of consciousness within a physicalist framework) is assumed by virtually all institutional funding, not because alternative frameworks have been empirically ruled out, but because the funding infrastructure was built by researchers within the dominant paradigm.
NSF and NIH cognitive neuroscience programs are organized around the assumption that consciousness is a product of neural computation. Grant proposals that treat this as the hypothesis to be tested — rather than the background assumption — face peer review panels constituted of researchers for whom the assumption is foundational. More critically, the instruments available for funding (fMRI, EEG, single-unit recording, optogenetics) are all neural measurement technologies. They are exquisitely sensitive to the neural correlates of conscious states. They are structurally unable to distinguish between "consciousness is what neural activity produces" and "neural activity is correlated with consciousness for reasons that don't establish the causal direction."
The funding consequences are documented in grant outcomes. Proposals in the integrative information theory (IIT) framework, the global workspace theory (GWT) framework, or quantum approaches to consciousness all require, at minimum, willingness to let the data challenge the materialist assumption. Proposals in these areas consistently receive lower priority scores than proposals that take the neural correlates approach — not because the results of the alternative approaches are known to be wrong, but because peer reviewers trained in neural correlates methodology evaluate methodological rigor according to criteria calibrated to neural correlates methodology. The alternative approaches look less rigorous because they are less consistent with the established instrument's epistemological standards. The standards are the instrument.
Researchers in consciousness science who pursue non-paradigm-aligned questions rely disproportionately on private foundation funding (Templeton Foundation, IONS, various private donors) rather than federal funding. This creates a two-tier system: paradigm-aligned research is supported by peer-reviewed federal grants, which confer academic legitimacy. Paradigm-challenging research is supported by private foundations, which confer the stigma of heterodoxy in the peer review environment. The funding architecture does not merely constrain what can be studied. It constrains who is treated as scientifically credible.
The Natural Selection of Bad Science
Smaldino and McElreath (2016) formalized the evolutionary dynamics that the above domain studies document empirically. They modeled scientific research as a population of "labs" — research units with varying methodological practices — that reproduce by producing publications. Labs with higher publication rates expand; labs with lower publication rates contract and eventually disappear. The key variable is not methodological rigor but publication rate.
Within their model, low-powered studies (those with insufficient sample sizes to reliably detect effects) have a selective advantage over high-powered studies. They are cheaper, faster, and — crucially — produce positive results more easily through the same flexible analysis practices documented in Section III. Labs that use low-powered methods with flexible analysis produce more publications per unit of research effort than labs that use rigorous methods. Under selection for publication rate, the population of labs shifts toward low-powered, flexible-analysis practices over time. The model produces replication crisis conditions as a stable evolutionary outcome, not as a deviation from normal science.
The Smaldino-McElreath model applies directly to instrument capture: labs that use established instruments with established analytical pipelines produce results faster, with higher publication rates, and with better citation outcomes than labs that invest in instrument development. The development cost is paid up front; the publication return is deferred and uncertain. Under selection for publication rate, the population of labs shifts away from instrument development toward instrument exploitation. Capture is the evolutionary stable strategy.
Smaldino and McElreath are careful to note that this dynamic does not require any individual scientist to behave dishonestly or negligently. The selection operates on career outcomes — grants, publications, hiring, promotion — not on intentions. Scientists who rigorously develop new instruments do not survive in the career environment that selects for publication rate. The ones who remain are those who successfully exploited existing instruments to produce publications. The incompetence is engineered not through deliberate design but through the emergent properties of a selection environment that rewards the wrong outputs.
The Ioannidis Audit
John Ioannidis's 2005 paper "Why Most Published Research Findings Are False" is the most cited paper in PLOS Medicine history. Its argument was statistical: under conditions of low study power, researcher flexibility in analysis and outcome selection, and publication bias for positive results, mathematical modeling shows that the majority of published research findings will be false positives — results that appear in publications but do not reflect real effects in the world.
The conditions Ioannidis identified are precisely the conditions produced by the incentive architecture described in this paper. Low power follows from the economic pressure to run cheap studies that maximize publication rate. Researcher flexibility follows from the career incentive to produce publishable results. Publication bias follows from journal preferences that select for positive outcomes. The Ioannidis result is not an independent critique of scientific practice. It is the mathematical expression of the incentive architecture's consequences.
Subsequent empirical validation has been extensive. The Open Science Collaboration (2015) attempted to replicate 100 published psychology findings and found that only 36% replicated at the original effect size. Prinz, Schlange, and Asadullah (2011) reported that Bayer HealthCare could replicate only 25% of published preclinical findings in cancer, cardiovascular, and women's health research. Begley and Ellis (2012) reported a 10% replication rate for 53 landmark cancer biology papers. The replication rates vary by field and methodology, but the direction is consistent: a substantial fraction of the published scientific literature — the output of the incentive architecture — does not survive independent verification.
The replication crisis data has a specific implication for instrument capture: the findings that fail to replicate were produced by established instruments using established methods. If those instruments were producing reliable knowledge, replication rates would be high. The low replication rates indicate that the established instruments — DSM-aligned assessments, low-powered cognitive neuroscience paradigms, in vitro cancer models — are systematically producing false positives. The captured field is not merely ignoring better questions. It is actively generating false answers to the questions it is addressing. Instrument capture does not just limit science. It corrupts it.
The Self-Perpetuation Loop
The incentive architecture that produces instrument capture is self-perpetuating through three feedback mechanisms that together make internal correction nearly impossible.
The personnel loop. The researchers who survive the career selection environment described above — those who successfully use established instruments to produce publications — become the senior researchers who sit on funding panels, journal editorial boards, and hiring committees. They evaluate proposals, papers, and candidates using criteria calibrated to the practices that produced their own success. Each generation of evaluators selects the next generation of researchers for instrument alignment, because instrument alignment was the survival strategy that produced the evaluators. The selection mechanism reproduces itself through personnel.
The infrastructure loop. Each generation of captured research produces findings that justify further investment in the instruments that produced them. Positive results from the dominant instrument justify expanding the infrastructure for that instrument. Expanded infrastructure increases the sunk cost of maintaining the captured paradigm, raises the threshold for the anomaly burden that would justify paradigm shift, and increases the number of researchers whose careers are indexed to the instrument's continuation. The infrastructure grows in proportion to the success of the capture.
The prestige loop. The dominant instrument accrues prestige through the citation economy: the most-cited papers use it, the highest-impact journals publish results derived from it, the Nobel prizes and national academy memberships go to researchers who advanced it. Prestige becomes evidence of the instrument's validity — not because prestige tracks truth, but because prestige tracks the metric performance the incentive architecture rewards. The instrument's dominance becomes self-confirming: the best scientists (as identified by the prestige hierarchy) use this instrument; this instrument is therefore what good science uses; proposals to use different instruments are judged by reference to this standard.
These three loops — personnel, infrastructure, and prestige — explain why instrument capture, once established, is not self-correcting. The Kuhnian revolution, in which anomalies accumulate until a paradigm shift becomes inevitable, requires that the anomalies reach a threshold within the scientific community's attention. The incentive architecture prevents that threshold from being reached by ensuring that researchers have stronger career incentives to reinterpret anomalies as measurement error than to develop the alternative instruments that would establish the anomalies as genuine. The captured equilibrium is stable not despite the scientists' intelligence but because of it: they correctly perceive that the costs of challenging the dominant instrument exceed the career rewards.
This is the architecture that Engineered Incompetence documents across three series: the accountability absence that removes oversight of the architecture itself; the instrument capture that removes the mechanisms of self-correction from within the scientific enterprise; and the workforce crisis that removes the researchers who would be most capable of recognizing and challenging the captured state. The three collapses share one root — the removal of productive friction — but nowhere is productive friction more systematically removed than in the incentive architecture of contemporary science.
The Legal Record of Incentive Capture — Three Adjudicated Cases
The argument of this series is structural: funding incentives shape research outcomes at the institutional level in ways that individual integrity cannot prevent. The following three cases move this argument from structural to adjudicated. In each, courts or regulators found — or settlements established — that research outcomes were systematically distorted by the incentive architecture of the institution producing the research. These are not allegations. They are the legal record.
In 2001, GlaxoSmithKline published Study 329 in the Journal of the American Academy of Child and Adolescent Psychiatry, concluding that paroxetine (Paxil) was “generally well tolerated and effective” for adolescent depression. The paper was largely ghostwritten by a medical communications firm retained by GSK. The underlying trial data, obtained through litigation and reanalyzed by Le Noury et al. in the BMJ (2015), showed that paroxetine did not outperform placebo on any primary efficacy endpoint and was associated with increased suicidal ideation in adolescents — the opposite of the published conclusions on safety. The instrument (the RCT) produced the right data. The incentive architecture (GSK’s $526M annual Paxil revenue at the time of publication) produced a published paper that reported the opposite of what the data showed. GSK pleaded guilty in 2012. The paper was not retracted until 2015 — fourteen years after publication, during which it was cited as evidence of paroxetine’s efficacy in adolescents.
The VIGOR trial, published in the New England Journal of Medicine in 2000, compared Vioxx (rofecoxib) to naproxen for gastrointestinal safety. The published paper reported that Vioxx produced significantly fewer GI adverse events. What the paper did not report was that the Vioxx group experienced significantly more serious cardiovascular events — specifically, that three myocardial infarctions that occurred in the Vioxx arm before the data cutoff were omitted from the published cardiovascular safety analysis. An FDA memorandum prepared in 2004 concluded that had the three omitted events been included, the cardiovascular risk ratio would have been statistically significant at the time of publication. Vioxx was withdrawn in 2004 after a separate trial confirmed cardiovascular risk; Merck agreed to pay $4.85 billion in 2007. The Vioxx case is the clearest documented instance of the Instrument Capture pattern at the clinical trial level: the instrument (the VIGOR RCT) generated data that included the cardiovascular signal. The incentive architecture (Merck’s $2.5B annual Vioxx revenue) determined which data reached the published paper.
Internal documents from Brown & Williamson Tobacco Corporation, released through litigation in 1994, established what the documents themselves stated plainly: the tobacco industry’s research strategy was not designed to resolve the question of whether smoking caused cancer. It was designed to sustain the appearance of scientific controversy. A 1969 Brown & Williamson document stated: “Doubt is our product since it is the best means of competing with the ‘body of fact’ that exists in the mind of the general public.” Industry-funded research was designed to produce uncertainty, not resolution. The research program was successful: it extended the period of regulatory inaction by decades. The Master Settlement Agreement of 1998 required the industry to release all research documents and established that the research program constituted consumer fraud. The tobacco documents are the founding archival record of manufactured scientific doubt as a deliberate institutional strategy — the upstream template for the incentive capture mechanisms this series documents in pharmaceutical, physics, and consciousness research.
Primary References
- Smaldino, P.E., & McElreath, R. (2016). The natural selection of bad science. Royal Society Open Science, 3(9), 160384. [Evolutionary model showing how publication selection produces low-reliability science as a stable equilibrium.]
- Ioannidis, J.P.A. (2005). Why most published research findings are false. PLOS Medicine, 2(8), e124. [Statistical demonstration that standard research conditions produce majority false positive findings.]
- Open Science Collaboration. (2015). Estimating the reproducibility of psychological science. Science, 349(6251), aac4716. [100-paper replication attempt; 36% replicate at original effect size.]
- Prinz, F., Schlange, T., & Asadullah, K. (2011). Believe it or not: How much can we rely on published data on potential drug targets? Nature Reviews Drug Discovery, 10(9), 712. [25% replication rate in Bayer preclinical research.]
- Begley, C.G., & Ellis, L.M. (2012). Raise standards for preclinical cancer research. Nature, 483(7391), 531–533. [10% replication rate for landmark cancer biology papers.]
- Fanelli, D. (2012). Negative results are disappearing from most disciplines and countries. Scientometrics, 90(3), 891–904. [Documents systematic increase in positive result rates across fields, inconsistent with genuine scientific progress.]
- Simmons, J.P., Nelson, L.D., & Simonsohn, U. (2011). False-positive psychology: Undisclosed flexibility in data collection and analysis allows presenting anything as significant. Psychological Science, 22(11), 1359–1366. [Demonstrates that legitimate analytic flexibility can produce false positive rates exceeding 60%.]
- Cosgrove, L., Krimsky, S., Vijayaraghavan, M., & Schneider, L. (2006). Financial ties between DSM-IV panel members and the pharmaceutical industry. Psychotherapy and Psychosomatics, 75(3), 154–160. [Documents 56% financial relationship rate among DSM-IV panel members.]
- Kuhn, T.S. (1962). The Structure of Scientific Revolutions. University of Chicago Press. [Standard reference for paradigm entrenchment and the conditions of scientific revolution.]
- Smolin, L. (2006). The Trouble with Physics. Houghton Mifflin. [Documents how publication and career incentives produced string theory's dominance in theoretical physics independent of empirical productivity.]
- Insel, T., Cuthbert, B., Garvey, M., et al. (2010). Research domain criteria (RDoC): Toward a new classification framework for research on mental disorders. American Journal of Psychiatry, 167(7), 748–751. [NIMH's attempt to develop an alternative to DSM-category-dependent research; the obstacles this effort encountered document the funding architecture described in this paper.]
- Button, K.S., Ioannidis, J.P.A., Mokrysz, C., et al. (2013). Power failure: Why small sample size undermines the reliability of neuroscience. Nature Reviews Neuroscience, 14(5), 365–376. [Documents systematic underpowering in neuroscience as a product of the incentive environment.]
- Muller, J.Z. (2018). The Tyranny of Metrics. Princeton University Press. [Comprehensive account of metric governance dysfunction across domains; contextualizes publication and grant metric systems.]
- United States v. GlaxoSmithKline LLC, No. 12-cr-10206 (D. Mass. July 2, 2012). $3B settlement. DOJ Press Release: "GlaxoSmithKline to Plead Guilty and Pay $3 Billion to Resolve Fraud Allegations."
- Le Noury, J. et al. (2015). "Restoring Study 329: Efficacy and harms of paroxetine and imipramine in treatment of major depression in adolescence." BMJ, 351, h4320. — Independent reanalysis using original patient-level data obtained through litigation.
- Topol, E.J. (2004). "Failing the Public Health — Rofecoxib, Merck, and the FDA." New England Journal of Medicine, 351(17), 1707–1709. — Documentation of omitted cardiovascular events in VIGOR trial publication.
- Merck & Co. Settlement Agreement. (Nov. 9, 2007). $4.85 billion settlement of Vioxx personal injury claims.
- Glantz, S.A., Slade, J., Bero, L.A., Hanauer, P., & Barnes, D.E. (1996). The Cigarette Papers. University of California Press. — Analysis of Brown & Williamson documents released through litigation (1994).
- Master Settlement Agreement. (Nov. 23, 1998). Attorneys General of 46 states v. major tobacco companies. $246 billion settlement.