Jacob Stegenga

WE AIM HERE to outline a theory of evidence for use. More specifically we lay foundations for a guide for the use of evidence in predicting policy effectiveness in situ, a more comprehensive guide than current standard offerings, such as the Maryland rules in criminology, the weight of evidence scheme of the International Agency for Research on Cancer (IARC), or the US ‘What Works Clearinghouse’. The guide itself is meant to be well-grounded but at the same time to give practicable advice, that is, advice that can be used by policy-makers not expert in the natural and social sciences, assuming they are well-intentioned and have a reasonable but limited amount of time and resources available for searching out evidence and deliberating.

Robustness is a common platitude: hypotheses are better supported with evidence generated by multiple techniques that rely on different background assumptions. Robustness has been put to numerous epistemic tasks, including the demarcation of artifacts from real entities, countering the “experimenter’s regress,” and resolving evidential discordance. Despite the frequency of appeals to robustness, the notion itself has received scant critique. Arguments based on robustness can give incorrect conclusions. More worrying is that although robustness may be valuable in ideal evidential circumstances (i.e., when evidence is concordant), often when a variety of evidence is available from multiple techniques, the evidence is discordant. †To contact the author, please write to: Jacob Stegenga, Department of Philosophy, University of California, San Diego, 9500 Gilman Drive, La Jolla, CA 92093; e‐mail: [email protected].

Harms of medical interventions are systematically underestimated in clinical research. Numerous factors—conceptual, methodological, and social—contribute to this underestimation. I articulate the depth of such underestimation by describing these factors at the various stages of clinical research. Before any evidence is gathered, the ways harms are operationalized in clinical research contributes to their underestimation. Medical interventions are first tested in phase 1 ‘first in human’ trials, but evidence from these trials is rarely published, despite the fact that such trials provide the foundation for assessing the harm profile of medical interventions. If a medical intervention is deemed safe in a phase 1 trial, it is tested in larger phase 2 and 3 clinical trials. One way to think about the problem of underestimating harms is in terms of the statistical ‘power’ of a clinical trial—the ability of a trial to detect a difference of a certain effect size between the experimental group and the control group. Power is normally thought to be pertinent to detecting benefits of medical interventions. It is important, though, to distinguish between the ability of a trial to detect benefits and the ability of a trial to detect harms. I refer to the former as power-B and the latter as power-H. I identify several factors that maximize power-B by sacrificing powerH in phase 3 clinical trials. If a medical intervention is approved for general use, it is evaluated by phase 4 post-market surveillance. Phase 4 surveillance of harms further contributes to underestimating the harm profile of medical interventions. At every stage of clinical research the hunt for harms is shrouded in secrecy, which further contributes to the underestimation of the harm profiles of medical interventions.

Medical scientists employ ‘quality assessment tools’ (QATs) to measure the quality of evidence from clinical studies, especially randomized controlled trials (RCTs). These tools are designed to take into account various methodological details of clinical studies, including randomization, blinding, and other features of studies deemed relevant to minimizing bias and error. There are now dozens available. The various QATs on offer differ widely from each other, and second-order empirical studies show that QATs have low inter-rater reliability and low inter-tool reliability. This is an instance of a more general problem I call the underdetermination of evidential significance. Disagreements about the strength of a particular piece of evidence can be due to different—but in principle equally good—weightings of the fine-grained methodological features which constitute QATs.

Measuring the effectiveness of medical interventions faces three epistemological challenges: the choice of good measuring instruments, the use of appropriate analytic measures, and the use of a reliable method of extrapolating measures from an experimental context to a more general context. In practice each of these challenges contributes to overestimating the effectiveness of medical interventions. These challenges suggest the need for corrective normative principles. The instruments employed in clinical research should measure patient-relevant and disease-specific parameters, and should not be sensitive to parameters that are only indirectly relevant. Effectiveness always should be measured and reported in absolute terms (using measures such as 'absolute risk reduction'), and only sometimes should effectiveness also be measured and reported in relative terms (using measures such as 'relative risk reduction')-employment of relative measures promotes an informal fallacy akin to the base-rate fallacy, which can be exploited to exaggerate claims of effectiveness. Finally, extrapolating from research settings to clinical settings should more rigorously take into account possible ways in which the intervention in question can fail to be effective in a target population.

I describe two traditions of philosophical accounts of evidence: one characterizes the notion in terms of signs of success, the other characterizes the notion in terms of conditions of success. The best examples of the former rely on the probability calculus, and have the virtues of generality and theoretical simplicity. The best examples of the latter describe the features of evidence which scientists appeal to in practice, which include general features of methods, such as quality and relevance, and general features of evidence, such as patterns in data, concordance with other evidence, and believability of the evidence. Two infamous episodes from biomedical research help to illustrate these features. Philosophical characterization of these latter features—conditions of success—has the virtue of potential relevance to, and descriptive accuracy of, practices of experimental scientists

A platitude that took hold with Kuhn is that there can be several equally good ways of balancing theoretical virtues for theory choice. Okasha recently modelled theory choice using technical apparatus from the domain of social choice: famously, Arrow showed that no method of social choice can jointly satisfy four desiderata, and each of the desiderata in social choice has an analogue in theory choice. Okasha suggested that one can avoid the Arrow analogue for theory choice by employing a strategy used by Sen in social choice, namely, to enhance the information made available to the choice algorithms. I argue here that, despite Okasha’s claims to the contrary, the information-enhancing strategy is not compelling in the domain of theory choice

Consensus conferences are social techniques which involve bringing together a group of scientific experts, and sometimes also non-experts, in order to increase the public role in science and related policy, to amalgamate diverse and often contradictory evidence for a hypothesis of interest, and to achieve scientific consensus or at least the appearance of consensus among scientists. For consensus conferences that set out to amalgamate evidence, I propose three desiderata: Inclusivity, Constraint, and Evidential Complexity. Two examples suggest that consensus conferences can readily satisfy Inclusivity and Evidential Complexity, but consensus conferences do not as easily satisfy Constraint. I end by discussing the relation between social inclusivity and the three desiderata

Amalgamating evidence of different kinds for the same hypothesis into an overall confirmation is analogous, I argue, to amalgamating individuals’ preferences into a group preference. The latter faces well-known impossibility theorems, most famously “Arrow’s Theorem”. Once the analogy between amalgamating evidence and amalgamating preferences is tight, it is obvious that amalgamating evidence might face a theorem similar to Arrow’s. I prove that this is so, and end by discussing the plausibility of the axioms required for the theorem