LL2-27 — Ch27 More or less precaution?#

Late lessons from early warnings: science, precaution, innovation (EEA Report No 1/2013), Part “Implications for science and governance”, Chapter 27. Report pages 643–669 (chapter text pp. 643–662; references pp. 662–669). PDF pages 645–671.

Read in full from the text extract (last page marker seen: PDF 671 / report p. 669). Checked visually against the PDF: summary box (p. 643), Box 27.2 (p. 650), Box 27.4 (p. 653), Figure 27.1 (p. 654), Table 27.1 (p. 656), Table 27.2 (p. 658), Figure 27.2 (p. 660), Box 27.5 (p. 661). The extract was accurate for these; no material garbling found. The front matter and Annex 1 were checked only for the author’s stated role.

Authors and standpoint#

Author: David Gee (sole author). The chapter does not state his affiliation. The report’s acknowledgements (PDF p. 7) say: “Originator of the Late lessons from early warnings project at the EEA was David Gee”, and list him first on the editorial team. That team also included Philippe Grandjean, Malcolm MacGarvin and Sybille van den Hove, whose work this chapter leans on (Ch 26; “the sin of hubris”; innovation “with a human purpose”). Annex 1 (PDF p. 691) describes him as a graduate in economics and politics. Since 1974 he had worked on occupational and environmental health “for trade unions and non-governmental organisations”. He was a former Director of Friends of the Earth UK and had been with the EEA since 1995 on “science, policy and emerging issues”. He was also “the catalyst, EEA editor and a chapters author” for both volumes. The table of contents (PDF pp. 5–6) shows he co-authored several of the case chapters this synthesis draws on: Ch 3 leaded petrol (with Needleman), Ch 8 vinyl chloride and Ch 21 mobile phones (with Hardell and Carlberg). So this is the project lead’s closing synthesis, not an independent review of the case chapters.

Evident stance. The chapter openly argues for wider use of the precautionary principle (PP). The summary box frames opposition in two ways. First, as “vested interests who perceive short term economic costs”. Second, as “intellectual resistance” from scientists who are “excessively attached to conventional scientific paradigms” and wait for very high strengths of evidence (p. 643). The title poses a two-sided question, but the chapter answers only one side (“more”). The chapter also works as the volume’s definitional and methodological statement: an EEA working definition of the PP, a typology of knowledge states, a strength-of-evidence scale, “criteria for action”, and a participatory risk-analysis framework.

Panels and commentaries. This chapter has no panels. There is one guest box: - Box 27.5 “Responsible research and innovation” (p. 661): “Edited extracts from von Schomberg, 2013”. René von Schomberg was at the European Commission’s DG Research, and a footnote says the views are his own and not an official Commission position (p. 661, fn 6). The box agrees with the chapter’s direction and does not dissent. Keep its claims (the “steerless” innovation critique, co-responsibility ethics, the state’s responsibility for positive outcomes) attributed to von Schomberg, not Gee. - Box 27.1 (pp. 648–649) compiles legal texts. Box 27.2 (p. 650) is Gee’s text built around a quotation from Barouki et al. (2012, misspelt “Baruoki”). Box 27.3 (p. 652) lists Bradford Hill’s nine features. Box 27.4 (p. 653) is Gee’s “criteria for precautionary action”. - Other voices quoted in the text: Jasanoff; Clair Patterson (1966 testimony, via Ch 3); Tee Guidotti (via Ch 6); Le Menestrel and Rode (Ch 25); Nicholas Stern (via Ch 14); Robert Costanza (via Ch 24); Austin Bradford Hill (1965); Needleman; Bellinger; Weiss and Bellinger; Lewontin (via Orrell); IUCN (1980); de Sadeleer; MacGarvin; the Interphone study group. Le Menestrel and Rode’s caution against “blaming business … with hindsight” is the main dissenting-in-tone voice, and Gee reports it fairly (p. 647).

Section-by-section notes#

Summary box (p. 643)#

• The PP is now in “a growing body of EU and national legislation and case law” but has been “strongly opposed by vested interests who perceive short term economic costs from its use” (p. 643). Scientific resistance is attributed to failure to acknowledge ignorance and uncertainty, attachment to paradigms, and waiting for very high strengths of evidence.
• Promised topics: confusing definitions (of the PP, prevention, risk, uncertainty, variability, ignorance); “common myths about the meaning of the precautionary principle”; handling of complexity and uncertainty; different strengths of evidence for different purposes. Note: no section is given over to “myths”. Misreadings come up piecemeal. Section 27.3 is titled “key elements and misunderstandings”, but it mostly sets out definitions. The one misreading of the PP’s meaning that the chapter rebuts explicitly is the reading of the North Sea Declaration (p. 657). Its other corrections concern related concepts rather than the PP itself: uncertainty vs ignorance (p. 654) and “no evidence of harm” vs “evidence of no harm” (p. 658).
• New concept: the “knowledge to ignorance” ratio. The PP “is particularly relevant where the ratio of knowledge to ignorance is low, as with emerging technologies” (p. 643).
• A working definition is offered to fix other definitions’ “triple negatives”, their silence on context, their lack of case-specific strengths of evidence, and their “overly narrow interpretations of the pros and cons of action or inaction” (p. 643).
• It calls for public engagement in “framing and decision‑making about both upstream innovations and their downstream hazards”, including specifying the EU treaty’s “high level of protection”. It proposes a precautionary, participatory framework for risk analysis, “along with some ‘criteria for action’ to complement criteria for causation” (p. 643).
• The capacity to “foresee and forestall disasters” is limited, especially against powerful interests. With humility and engagement, societies could use the PP to minimise hazards “whilst stimulating innovation”, and get “more socially relevant and diverse innovations” (p. 643).

27.1 Introduction (pp. 644–645)#

• Epigraph: de Sadeleer (2010), the PP “has, within the space of a decade, experienced a meteoric rise” (p. 644).
• Since Volume 1 (2001) the PP has entered “many laws and constitutions”. France constitutionalised it in 2005 (p. 644).
• The PP as a trigger for wider debate. The French GMO debate of 1997–2005 (Marris 2005) moved “from narrow questions of risk and scientific uncertainty to broader questions about the future of agriculture, the direction of scientific research and innovation, and public engagement” (p. 644). Stirling (2008): opening up debate can empower “wider social agency in technology choice”. De Marchi (2003): such debates clarify “what [the] conflict is really about” (p. 644). Ch 19 (GM crops and agro-ecology) is cited: “bottom up” approaches “are proving capable of getting sustainable, participatory and locally adapted solutions into the hands of those that need them most” (p. 644).
• Two roles of the PP (p. 644): (1) a trigger for broad debates on “what kind of future we want” and which innovation pathways lead there; (2) “a legal and moral justification for more timely actions on early warnings”. The volume emphasises role 2. Parts B and C “begin to illustrate” role 1, and Chs 24 and 26 illustrate reform of science, law and scientific organisations (p. 644).
• Case material recapped (p. 644):
• Known hazards with early warnings and “(usually) late actions”: asbestos, benzene, BSE, DES, TBT and PCBs (Vol. 1); DDT, DBCP, VCM, lead in petrol, mercury, beryllium and booster biocides (Vol. 2).
• These known-hazard cases “primarily illustrate how more precautionary action could be applied to chemical risks emerging now”. The examples given are BPA and other chemicals, “nicotinoid pesticides”, and endocrine disrupters in consumer products, including ethinyl oestradiol “in the pregnancy pill” (Ch 13). At p. 649 the same pill is called “the contraceptive pill”, so “pregnancy pill” is a slip.
• Technology histories (X-rays, fishing techniques, fossil fuels, early nuclear) are presented as lessons for “such emerging technologies as nanotechnology, genetically modified (GM) food, radio‑frequency from mobile phones, and the new generation of nuclear plants” (p. 644).
• The chapters on alien species, floods and ecosystems, and the late actions on climate change, “provide insights into how the management of ecosystems could develop” (p. 644).
• High cost of inaction. Late action on known hazards “illustrate[s] the high cost of inaction”. Globally that cost “has been paid in millions of lives and cases of disease and dysfunction, much damage to the environment and species, and very large economic penalties” (p. 644; cross-ref Ch 23). No figures are given here.
• Precautionary actions: five of 34 case studies (pp. 644–645). The chapter’s words are “describe precautionary actions”. It does not call them “successes”, but says they “illustrate the value of the PP”:
• the EU ban on hormones in cattle feed;
• EU regulation and some member-state actions on GMOs;
• France’s TBT ban “in 1984”;
• France’s Gaucho ban in 1999;
• “arguably, the belated but still precautionary” EU ban on antibiotics as growth promoters. These, with “the histories of the 88 claimed false positives analysed in this volume”, “illustrate the value of the PP in minimising harm and societal costs” (p. 645). The false-positives finding itself (from Ch 2, Hansen and Tickner) is not stated here, so this sentence cannot be checked from this chapter alone.
• Rapid action once compelling human evidence appeared. In other cases action “was taken quite quickly but only after serious and compelling human evidence became available” (p. 645). So these are examples of fast prevention, not of precaution. In DES and VCM the evidence came from “an observant clinician”; in DBCP it came from “the victims themselves”. “Just four to seven cases of very rare cancers or sperm reduction (DBCP) were sufficient to justify prompt regulatory action” (p. 645).
• Seven barriers to wider use of the PP (p. 645): 1. opposition from powerful corporations, “supported by some scientists, policymakers and politicians”, fearing economic, intellectual and political costs; 2. misunderstandings of the PP’s definition; 3. difficulties with complex systems “characterised by multi‑causality, scientific uncertainty, ignorance and scientific ‘surprises’”; 4. tension between the high strength of evidence for scientific causality and the lower strength needed for timely policy; 5. inadequate analysis of the costs and benefits of action and inaction, and “unrealistic market prices for hazardous agents”; 6. “political and financial short‑termism”; 7. “a failure in most cases” to engage civil society and the public “to help counter the power of the corporate and other stakeholders that may wish to dismiss early warnings”. “Taken together, these barriers explain much about the decades‑long delay between warnings and action” (p. 645). Barriers 1, 6 and 7 concern “political and economic power”; barriers 2–5 concern “the more technical process of applying knowledge to policymaking” (p. 645). The chapter adds that the two sides cannot be cleanly separated.
• Jasanoff (1990, 2011): “Policy relevant science comes into being in a territory of its own that is subject to neither purely scientific nor wholly political rules of the game”. Also: “It is not so much scientists (but) experts, who govern the production and evaluation of policy relevant science” (p. 645).
• Plan of the chapter (p. 645): corporate power (27.2); technical barriers 2–5 organised by the EEA definition (27.3–27.6); short-termism (27.7); public engagement (27.8).

27.2 The power of corporations to oppose action (pp. 645–647)#

• Opens with Carson’s line (via Ch 11 on DDT) that corporations focus on “making a dollar at whatever the costs”. Gee concedes “this overstates the situation”, but says the cases give “ample evidence” of “product defence” campaigns (p. 645). Ch 7 on environmental tobacco smoke identifies seven strategies. The tobacco industry “was certainly not alone”: lead, VCM, beryllium and climate change are named (p. 645).
• Prediction: “it seems likely that other industries with hazardous products to defend today would employ similar strategies, including trying to control, directly or indirectly, the relevant scientific research” (pp. 645–646).
• Control of research. The leaded petrol industry “maintained a virtual monopoly on leaded petrol research” from the 1925 “one day trial”, when senior public health scientists raised early warnings, “until the 1970s” (p. 646). “Without access to independent research the regulatory authorities were vulnerable to corporate influence on the scientific evidence made available to them” (p. 646). Clair Patterson told Congress (Muskie hearings, 1966) that public health agencies collaborating with industries to decide whether health is endangered is “a direct abrogation and violation of the duties and responsibilities of those public health organisations” (p. 646, fn 1).
• Research agendas now. They are “often determined by more independent academics and public sector organisations”. But framing “can result in research that focuses much more on developing products than on the need to find out whether those products are harmful” (p. 646). Evidence:
• EU public research funding over “the past two decades” on nanotechnology, biotechnology and information technology put “only about 3 % of the EUR 28.5 billion budget” into hazards (p. 646). No source is cited for this figure.
• USDA, 1992–2002: USD 1.8 billion on biotechnology research, “just 1 %” risk-related (Mellon 2003, via Ch 19) (p. 646).
• Intellectual property and access. For GMOs, access to the organisms themselves was restricted. There was “some opening up” after a complaint from 26 US academics whose research was inhibited by lack of access to corporate-owned seeds (Pollack 2009) (p. 646).
• Funding of innovation pathways. The Commission’s Standing Committee on Agricultural Research (SCAR 2012) calls for top priority for “low‑input high‑output systems” that “integrate historical knowledge and agro‑ecological principles that use nature’s capacity” (p. 646).
• Bias toward known problems. Environmental science has “a strong bias … towards research on well known problems rather than on emerging issues” (p. 646; Ch 26).
• Language. Corporations “have also realised that the language used in debates … is also important” (p. 646):
• Leaded petrol: “normal” blood lead levels framed as “natural” and therefore safe.
• “Sound science”: a term taken over by tobacco PR to mean science supporting the industry position (Baba et al. 2005), then used by other industries to call opposing science “unsound” (p. 646).
• “Manufacturing doubt” out of scientific uncertainty was “a key part of product defence” in the tobacco, lead, asbestos, beryllium, benzene and climate change cases (Michaels 2008; Oreskes and Conway 2010) (p. 646).
• Why corporations do it. “The long history of corporate misconduct begs the question why” (p. 646). Guidotti (Ch 6, beryllium): corporations lock into product defence because of “fear, denial and risk of loss”. If they are to reverse course, “there must be room for them to turn around”. His suggestion of “forgiving past liabilities and reducing punitive damages” “will be controversial”, while shareholder engagement “is likely to be welcomed” (p. 646).
• Ch 25, Le Menestrel and Rode (p. 647):
• “blaming business, in particular with hindsight … may not always be constructive”, because it misses a “complex or even contradictory set of motives and drivers”.
• Corporate decisions mix economic, epistemological, regulatory, cultural and psychological factors, and economic motives dominate: “in virtually all reviewed cases from both volumes … it was perceived to be profitable for industries to continue using potentially harmful products or operations” (p. 647).
• Externalisation. “Corporate short‑term interests have dominated over longer‑term public interests mainly because the costs of damage to people and environments were, and still are, largely externalised to society as a whole” (p. 647). Stern calls climate change “the biggest market failure ever” (via Ch 14). Even when victims win compensation, “the sums may be largely covered by insurance” (p. 647).
• Remedies. Internalise costs “via regulations, taxes, charges and permits” (Ch 23), plus “anticipatory assurance bonds” (Costanza on Deepwater Horizon, Ch 24) (p. 647). Le Menestrel and Rode propose:
• distinguishing the “economic” and “political” roles of businesses that have “ample opportunity to influence the regulatory process” (Scherer and Palazzo 2011; UCS 2012);
• new institutions for “rigorous and explicit exposition of the dilemmas and trade‑offs” and of “the organisational pressure to deny the reality of the early warnings”. Such institutions would “more realistically complement initiatives based on the idealised principle that being socially responsible is economically profitable” (p. 647).
• Responsible corporate behaviour. The historical cases reveal only “one or two examples”, “albeit by companies selling hazardous products rather than by their manufacturers” (p. 647). That qualifier applies to the historical examples (asbestos users, CFC aerosols) and to the BPA users. Several of the more recent examples are manufacturers (BASF, Hewlett Packard, AstraZeneca). The list:
• some companies stopped using asbestos in the 1970s;
• “Johnson & Johnson stopped using CFCs in their aerosols in 1977, eight years before the ozone hole was discovered” (see errors below);
• some user companies dropped BPA before the Commission acted “on its use in baby toys” (see errors below);
• the marine and forest stewardship councils;
• BASF working with civil society on nanotechnology codes of conduct (EU 2010);
• Hewlett Packard removing lead and other hazardous compounds from electronic goods;
• AstraZeneca researching “green” medicines;
• Puma on environmental accounting;
• the WBCSD’s business vision (p. 647).

27.3 The precautionary principle: key elements and misunderstandings (pp. 647–649)#

• Moving from “evidence to action” is “a balancing act between what needs to be known and what ought to be done” (Weed 2004) (p. 647).
• Tobacco as the paradigm case of lost precaution. “More than 40 years” of science and debate, from the 1940s to the 1980s, passed before knowledge about smoking and lung cancer protected public health, “following sustained opposition from economic and political interests”. “The opportunity for precautionary action on a likely hazard in the 1950s and 1960s was lost. By the 1990s only prevention of known harm was possible” (p. 647).
• Definitions vary on: the standard of evidence needed to invoke the PP; the obligation on public bodies; the objectives; and whether provisions on costs and benefits or public participation are included (pp. 647–649).
• Box 27.1 (pp. 648–649) quotes: Rio Declaration 1992, Principle 15; TFEU Art. 191(2) (the PP plus “a high level of protection”); REACH preamble and Art. 69; UNFCCC 1992 Art. 3(3), including cost-effectiveness; Directive 2001/18/EC Art. 4(1); Cartagena Protocol 2000 Art. 11(10); Regulation 178/2002 Art. 7 (“provisional risk management measures … pending further scientific information”); Regulation 1107/2009 Art. 1(4); the IMO Anti-fouling Systems Convention (dated “2000” in the box); ECJ BSE case C‑157/96 (1998), where institutions “may take protective measures without having to wait until the reality and seriousness of those risks become fully apparent”; Stockholm Convention 2001; WTO SPS Art. 5(7); and the Commission Communication of 2 February 2000 (“reasonable grounds for concern”).
• EEA working definition (p. 649), quoted in two parts:
• “The precautionary principle provides justification for public policy and other actions in situations of scientific complexity, uncertainty and ignorance, where there may be a need to act in order to avoid, or reduce, potentially serious or irreversible threats to health and/or the environment,”
• “using an appropriate strength of scientific evidence, and taking into account the pros and cons of action and inaction and their distribution” (p. 649).
• Claimed advantages (p. 649):
• it names situations of uncertainty, ignorance and risk as the contexts;
• it is “expressed in the affirmative rather than the triple negatives found in, for example, the Rio Declaration”;
• it acknowledges that the strength of evidence needed is “determined on a case‑specific basis, and only after the plausible pros and cons, including their distribution across groups, regions, and generations, have been assessed”. The chapter says it has “proved useful in helping to achieve a more common understanding” (p. 649). No evidence is offered for that uptake claim.
• The three sets of issues in the definition structure 27.4–27.6: complexity and uncertainty, strength of evidence, and pros and cons (p. 649).

27.4 Complex biological and ecological systems (pp. 649–656)#

• Reproductive and developmental hazards are the illustration. Cases showing such harm: Minamata mercury, TBT, DES, PCBs, tobacco, lead, VCM, ethinyl oestradiol, BPA and X-ray radiation (p. 649). Harm “can be initiated in the early life stages … but may not become apparent until much later in adult life, and even in subsequent generations, as in the DES case” (pp. 649–650).
• Box 27.2 (p. 650). Developmental periods are highly sensitive. Barouki et al. (2012) link a long list of rising conditions “in part” to developmental factors: obesity, diabetes, hypertension, cardiovascular disease, asthma and allergy, immune and autoimmune disease, neurodevelopmental and neurodegenerative disease, precocious puberty, infertility, some cancers, osteoporosis, depression, schizophrenia and sarcopenia. Mechanisms are “not yet well established, despite decades of research”. But “it seems clear that it is more the timing of the dose, rather than the dose itself” that distinguishes harmful from harmless exposure (p. 650). Such harm is “often irreversible and sometimes multigenerational” and cannot be offset by any benefit to the individual from intrauterine exposure. So “biology, economics, equity and morals all justify early actions” (p. 650). Evidence is much harder to establish than for tobacco and lung cancer, where one agent caused a specific, previously rare cancer (p. 650).
• “From monocausality to multicausality” (p. 650). Chronic disease processes “appear to involve some or all of at least eight main events”: preparation within the host, initiation, promotion, retardation, progression, disease onset, strengthening or weakening of severity, and prevalence. These are acted on by “many interdependent, co‑causal risk factors, where the timing of exposures is usually critical”, with chance also playing a part (p. 650). Exposures are “interactive, mixed, and usually low level” and affect people with specific histories and susceptibilities (p. 650).
• There is a long-standing tension between monocausal, reductionist approaches (“the germ, or the gene, the oncogene, or a single risk factor”) and multicausal, holistic approaches built on the host’s environmental history (Sing et al. 2004). Ecology has a parallel tension between “diversity” and “variable” approaches (Ch 17) (p. 650).
• “Concentrating research on particular parts of the puzzle, rather than on the causal puzzle itself, may inhibit the clarification of causality”. Tobacco smoke contains some 4,000 chemicals, over 100 classified toxic, yet why some smokers get cancer or heart disease and others do not “is still largely unknown after more than 40 years” (p. 650).
• Acting on whole mixtures works despite ignorance of mechanism. Reducing exposure to tobacco smoke, fossil-fuel smoke, welding and rubber fumes, and fine particles has prevented harm (p. 651).
• “Reductionism and the metaphor of the body as a machine are powerful paradigms”. They support linear cause–effect thinking “long after knowledge about irreducible uncertainties, emergent properties and non‑linear dynamics became available” (Di Giulio and Benson 2002) (p. 651).
• “From confounders to co-causal factors?” (p. 651).
• The tools for multicausal analysis are “not well developed or used” (Cory-Slechta 2005). Epidemiologists isolate single risk factors and statistically remove confounders. “Such ‘statistical surgery’ or ‘context stripping’ may remove many confounders from the analysis that are really co‑causal factors” (p. 651).
• Weiss and Bellinger (2006): treating environmental conditions as confounders “is equivalent to defining genetic differences as confounders” (p. 651).
• Lead: lead exposure, deprivation/enrichment and neurotoxicity are “perhaps bidirectional” (Bellinger 2007). An enriched home reduces lead’s harm, and deprivation increases it (p. 651).
• Cell signalling and crosstalk imply “causality may be circular” (Soto and Sonnenschein 2006) (p. 651).
• Ecotoxicology must infer across biological scales, for example from harm to individual fish to population decline (Kidd et al. 2007; Ch 13) (p. 651).
• Bellinger et al. (1985): interacting systems “will always overwhelm predictions of independent effects of any single factor”. In non-linear systems, “Small’ can be very significant”, and removing the “smallest” link can break a pathway to disease (p. 651).
• Bradford Hill (pp. 651–654). The “decisive question” is “whether the frequency of the undesirable event B will be influenced by a change in the environmental feature A” (Hill 1965, p. 651). Box 27.3 lists the nine features: strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment (reversibility) and analogy (p. 652). These “subsequently misnamed ‘criteria’” are still widely used (p. 652).
• Hill’s approach “was essentially based on monocausality”, that is, on finding “the specific cause of a specific disease”. Gee qualifies this at once. Hill “was aware that several factors would be implicated in disease” and that removing one might reduce frequency without eliminating the disease. He also acknowledged that one disease can have several independent causes: “It has always been possible to acquire a cancer of the scrotum without sweeping chimneys or taking to mule spinning in Lancashire” (p. 652).
• The criteria “need to be reappraised in the light of multicausality and complexity”, because “the absence of some or all ‘criteria’ is often used in current controversies to deny the possibility of causality”. They “seem less robust now as reasons for dismissing associations than they did in the world of the 1960s, when issues were perceived in largely monocausal terms” (p. 652).
• Consistency: hard to achieve under multicausality. Needleman (1995): “if all studies of lead showed the same relationship between variables, one would be startled, perhaps justifiably suspicious” (p. 652). Variability comes from framing, models, methods, statistics, population choice, susceptible subgroups and data selection (Bailar 2007; Grandjean 2008), plus “sociomics” and epigenetics. Lead studies “can still only ‘explain’ 30–40 % of the variance” (Bellinger 2007). So “the absence of consistency between studies does not imply an absence of causality” (p. 652).
• Dose-response: described as “a linear dose-response relationship” criterion. Where timing matters more than dose and “non‑linear, ‘low‑dose’ effects are present”, absence of linearity “does not provide robust evidence against causality” (p. 652).
• Temporality: with multiple causes rising and falling at different times, overall trends can mislead (p. 652). Example: a WHO (2002) review concluded that because sperm counts began falling before the rise of chlorine chemistry, such exposures “could not be a cause of change in the overall trend”. Gee says this “is not soundly based” under multicausality (p. 653).
• Experiment: removing one cause of a multicausal outcome (IQ, sperm counts) rarely shows clearly (p. 653).
• Specificity: deserves less weight given “many to many” relationships. Tobacco, PCBs, asbestos, lead and mercury each cause many harms (p. 653).
• Strength: still relevant. “Even a ‘low’ relative risk of say, 1.5, if replicated in several studies, can be very robust for a multicausal disease”, as with smoking and heart disease, and it means much harm if baseline rates are high (p. 653).
• Plausibility and coherence: “dependent on the established knowledge of the day”, so they are not robust grounds for dismissing early warnings from frontier science (p. 653).
• Analogy: “Whereas multicausality seems to weaken most of the criteria, analogy becomes more necessary” (p. 653). This leads into Box 27.4.
• Box 27.4 “Criteria for precautionary action” (p. 653), 12 features of evidence that “may justify precautionary action”: 1. intrinsic toxicity/ecotoxicity data; 2. novelty (“low ‘knowledge/ignorance ratio’”); 3. persistence; 4. bioaccumulation potential; 5. large spatial range / global dispersion; 6. seriousness; 7. irreversibility; 8. analogous evidence from known hazards; 9. inequitable distribution of impacts across regions, people and generations; 10. feasible alternatives; 11. “potential for stimulating innovation”; 12. potential and time scales for future learning. The text says these are “based on experiences of past ecological and biological hazards, including by analogy, which may provide quite robust evidence of emerging potential hazards” (p. 653). No weighting or decision rule is given.
• Statistical significance. Hill warned against its misuse: “we grasp the shadow and lose the substance … far too often we deduce “no difference” from “no significance”” (p. 653). Such cautions have been repeated (Cohen 1994; Poole 2000; Hooper, Stang and Rothman 2011), yet misinterpretation and “neglect of confidence intervals continue” (p. 653).
• Asymmetry. “The presence of the criteria can be robust evidence for a causal association, whereas the absence of the criteria is not robust evidence that there is no causal association” (p. 653). Hill noted this for several criteria, “but some of his followers have forgotten this in their use of the criteria to dismiss possible hazards (Ashby, 1997; WHO 2002)” (pp. 653–654). The asymmetry is “even more pronounced” under complexity (p. 654).
• Methods biased toward false negatives. There are “systemic biases towards not finding a causal link”, specifically within epidemiology and toxicology methods that generate false negatives, “assertions that something is safe when it turns out not to be” (Grandjean, Ch 26) (p. 654). Many scientists acknowledge this, “but awareness of these methodological biases among many stakeholders appears to be low” (p. 654).
• Funding bias. Research results “closely associated with the source of funding”, seen in tobacco (Barnes 1998), pharmaceuticals (Goldacre 2012; Lexchin 2003), food and beverages (Levine 2003), BPA (vom Saal 2005), mobile phones (Huss et al. 2007), biomedicine (Bekelman 2003) and GMOs (Diels 2011). “The explanation for this bias is not clear” (Krimsky 2006, 2010). It is also present in transport and construction (“underestimation of costs and construction times by the developers is routine”) and in “cost‑benefit analysis where the direction of bias is routinely in the direction of those who fund the study” (p. 654). The last two claims are unreferenced.
• Uncertainty vs ignorance (p. 654). Asbestos mesothelioma and the CFC ozone hole were “complete scientific ‘surprises’, arising from a state of ignorance”, not gaps in existing knowledge. “Uncertainty” and “gaps” relate to “a stock of existing knowledge”. “Ignorance” (“nescience”) relates to “unknown unknowns”. Research closes some gaps but “will also uncover new sources of uncertainty and gaps in knowledge, as well as raising awareness about new areas of ignorance”. “Learning to live with and manage irreducible uncertainties is as necessary as trying to reduce them” (p. 654).
• Figure 27.1 “Expanding knowledge, continuing uncertainties” (p. 654). Nested regions: today’s knowledge (with numbered gaps), tomorrow’s knowledge (with new gaps), and an outer zone of “No knowledge (ignorance) source of surprises and discoveries”, with arrows pushing outward. The point: as knowledge grows, its boundary with ignorance grows too.
• The knowledge-to-ignorance ratio (KIR) (pp. 654–655).
• IUCN’s World Conservation Strategy (1980): “what we know about the biosphere, ecosystems and their interrelationships is less than what we do not know … in the meantime risks should be reduced” (p. 655).
• This calls for “scientific humility”. “Scientists in many of the case studies failed to show scientific humility, instead being guilty of what has been termed ‘the sin of hubris’” (MacGarvin 1994, on fisheries) (p. 655). No specific cases are named here.
• The KIR “cannot be quantified” but can be appreciated qualitatively. It is high for asbestos, ionising radiation and tobacco, and low for “nanotechnology, GM food, or … non‑ionising radiations when used in recent consumer products such as mobile phones” (p. 655). For asbestos, “the practical need to search for more knowledge … is very minimal” (p. 655).
• Footnote 2: Dutch analyses (Lenters et al. 2012) found the claimed potency difference between blue and white asbestos “practically disappears” when exposure estimates are scrutinised (p. 655).
• “There is a vast ocean of scientific ignorance surrounding nanotechnologies, biotechnologies and non‑ionising radiation technologies and chemicals used in consumer and other products” (p. 655). Newton’s “pebbles” and “great ocean of truth” are invoked (fn 3).
• Rule of thumb: high KIR (lead, asbestos, mercury) means little need for more research or for precaution “(as distinct from merely preventative)” measures. Low KIR means a need for “both precautionary measures following credible early warnings and for novel research, rather than the ‘scientific inertia’ of excessive research on well known substances” (Ch 26) (p. 655).
• De Sadeleer (2010): full risk assessment may be impossible at the frontier, where scientists “must even point to the limits of their knowledge or … to their ignorance” (p. 655).
• “The limitations of scientific knowledge imply moral courage in taking precautionary action in time to avert harm” (p. 655). Lewontin (via Orrell 2007): “It takes a certain moral courage to accept the message of scientific ignorance and all that it implies” (p. 655).
• Table 27.1 “Some common concepts used in PP debates” (p. 656):

Note: “Different types, sources and levels of uncertainty can be identified (Walker, 2003)”. Walker 2003 is not in the reference list.

27.5 Conflicts between the high strength of evidence needed for scientific causality and the lower strength needed for timely public policy (pp. 655–658)#

• Minimum threshold: “All responsible applications of the precautionary principle require some plausible evidence of an association between exposures and potentially harmful impacts” (p. 655). The Commission’s 2000 Communication requires “reasonable grounds for concern”. But it does not explain that these grounds vary case by case, “nor does it explicitly distinguish between risk, uncertainty and ignorance” (pp. 655–656).
• Thresholds vary and “can be quite low” (p. 656). In Sweden a “scientific suspicion of risk” suffices to restrict an existing chemical. Under WTO SPS, “pertinent scientific information” suffices where science is insufficient for a full risk assessment (p. 656).
• What sets the appropriate strength: the pros and cons of action and inaction in the case. That means “the nature and distribution of potential or plausible harm; the justification for and the benefits of the agent …; the availability of feasible alternatives; and the overall goals of public policy”, including the EU treaty’s “high levels of protection” (p. 656).
• Precedent: criminal “beyond all reasonable doubt” vs civil “balance of probabilities” (p. 656). Bradford Hill’s 1965 closing “call for action” (pp. 656–657): “It almost inevitably leads us to introduce differential standards before we convict”. On “relatively slight evidence”, restrict a morning-sickness drug (“The good lady and the pharmaceutical industry will doubtless survive”). On “fair evidence”, switch to a non-carcinogenic oil at work. But “very strong evidence” is needed before making people give up a fuel, cigarettes, fats or sugar they like (emphasis added by Gee). IARC also grades evidence strengths (Cogliano 2007) (p. 657).
• Misreading of the North Sea Declaration (p. 657). Critics say its call for action “even where there is no scientific evidence to prove a causal link” justifies action with “no scientific evidence”. Gee answers that “no scientific evidence” is tied to “to prove a causal link”. There is “a significant difference” between evidence showing a plausible link and evidence robust enough to “prove” one. “Once evidence reaches the level of ‘proving’ a causal link there is no need for the PP as the issue is then firmly in the ‘prevention principle’ area” (p. 657).
• Interphone and the media (p. 657). The Interphone study was a 13-country study of brain cancers and mobile phones (Ch 21). Gee treats it as a “similar confusion” to the North Sea case. He says confusion among commentators, “including the media”, arose because “scientists were not transparent and clear about the difference between the very strong evidence needed to establish ‘causality’ and the suggestive evidence of plausible risks”. The chapter’s version of the conclusion is: “There were suggestions of an increased risk of glioma, and much less of menigioma, at the highest level of exposure….. (but) biases and errors limit the strength of the conclusion … and prevent a causal interpretation”. Readers took this as either no evidence of cancer or evidence of cancer. BBC, 17 May 2010: “No proof of mobile cancer risk, major study concludes”. The Telegraph, same day: “Half an hour of mobile use a day increases brain cancer risk” (p. 657). See the bias check: the published abstract’s conclusion opens with a sentence the chapter omits.
• The evidentiary continuum. “Beneath the strong evidence of ‘scientific causality’ there is a large evidentiary space containing a continuum of strengths of evidence” usable under the PP. “The question remains, however, where, in that continuum, is ‘sufficient evidence’ located?” (p. 657).
• IPCC as a model. “Balance of evidence” in 1995; seven strengths of evidence in 2001 (Panel 14.1); “very likely” with “high confidence” by 2007 (p. 657). Table 27.2 “presents five of these strengths of evidence based on the IPPC [sic] approach” (p. 657). So the percentage bands are adapted from the IPCC’s 2001 likelihood scale, not devised independently.
• Table 27.2 “Different strengths of evidence for different purposes” (p. 658):
• Very strong (90–99 %): “statistical significance”, part of strong evidence of causation; “beyond all reasonable doubt”, criminal law and the Swedish Chemical Law 1973 for evidence of “safety” of suspect substances, “placing the burden of proof on manufacturers”.
• Strong (65–90 %): “reasonably certain” (US Food Quality Protection Act 1996); “sufficient scientific evidence” (WTO SPS Art. 2) for a trade restriction.
• Moderate (33–65 %): “balance of evidence” (IPCC 1995, 2001); “balance of probabilities” (civil law); “reasonable grounds for concern” (EC Communication 2000); “strong possibility” (British Nuclear Fuels occupational radiation compensation scheme 1984, where 20–50 % probabilities trigger graded awards and 50 %+ triggers full compensation).
• Weak (10–33 %): “scientific suspicion of risk” (Swedish Chemical Law 1973 for precautionary action, “placing the burden of proof on the regulators”); “available pertinent information” (WTO SPS Art. 5.7 provisional restriction).
• Very weak (1–10 %): “low risk” (household fire insurance); “negligible and insignificant” (FQPA 1996).
• The same Swedish law appears at both ends: manufacturers need very strong evidence of safety, while regulators need only weak evidence of risk to act. This is a compact illustration of how burden and standard of proof can be split between parties.
• Beyond science. “The decision about when there is sufficient evidence to justify preventive action clearly involves more inputs to decision‑making than merely science”. The non-scientific inputs include the costs of being wrong and their distribution “between different groups and generations”, benefits, and alternatives (p. 657).
• The rhetoric of “not established.” “No established or conclusive evidence” usually means the evidence has not convinced the particular assessors (p. 657). “The different consequences for those for whom the evidence is ‘not established’ (i.e. risk takers or risk makers) is seldom discussed”. Nor is the purpose for which evidence would be conclusive: “warning labels, or low cost exposure reductions, or a ban” (p. 658).
• “No evidence of harm” vs “evidence of no harm”. Decision-makers must avoid assuming the former is the latter “when the relevant research has not been done”, a feature of many cases (Ch 26: “authoritative but unsubstantiated assertions of safety”) (p. 658).
• Risk assessors should consider consequences. Reading “convincing evidence” only as proof of causality “is of little practical use in helping to apply the precautionary principle or in averting, as opposed to observing, future harm” (p. 658). So risk assessment committees “may need to consider the consequences of their judgements as well as just causation”, as Hill and the IPCC did. Gee acknowledges the counter-view (“It may be argued that it is for risk managers to deal with the consequences”) He replies that scientists “are well placed to contribute to analysis of consequences”, and that the European Court wanted assessments to convey “the ramifications of the scientific question” (cited as “ECR, 1999 and 2002” (p. 658); the reference list identifies these as Case T‑13/99 Pfizer and Case T‑70/99 Alpharma). Both are Court of First Instance cases. From background knowledge, both judgments were given in 2002; “1999” is the year the case was lodged.

27.6 The pros and cons of actions and inactions (pp. 658–659)#

• The EEA definition widens “conventionally narrow and quantifiable” cost–benefit analysis to “wider and sometimes unquantifiable ‘pros and cons’” (p. 658). The key example is “a loss of trust in science after the public experiences harm that scientists had assured them would not occur”, which “can sometimes be as significant as the economic costs”: BSE, Chernobyl and Fukushima (Ch 18) (p. 658).
• Skewed accounting (Ch 23): costs and benefits are “skewed towards the tangible short‑term compliance costs of regulatory action, which usually fall on specific, often powerful actors, and against the long‑term diffuse benefits to society as a whole of timely actions” (p. 659). The polluter-pays principle and internalisation would bring prices into line with real costs, “encouraging earlier development of substitutes and other economic and technological innovations” (EEA 2012) (p. 659).
• Innovation and lock-in. Asbestos, lead, mercury, PCBs, CFCs and benzene “indicate that early actions can stimulate innovations”. Conversely, “late actions have consolidated technological monopolies for products, at unrealistically low prices, which served to keep smarter substitutes out of the markets for many years” (p. 659). Supporting literature on regulation-induced innovation: Porter 1995; Ambec 2011; Ashford 1979, 2011a, 2011b, 2012 (p. 659). No data is presented here.

27.7 Political and financial short‑termism (p. 659)#

• “The time horizons of democratic politics are very short in comparison to the long timescales associated with successfully managing the harm” (p. 659).
• Finland, Israel, New Zealand and Hungary “have been experimenting” with ombudsmen or committees for the long term (Roderick 2010; Ward 2012) (p. 659).
• The financial sector is “even more limited by short termism”. Since the crash there has been “some effort” toward longer-term perspectives (Mainelli and Giffords 2009) (p. 659). This is the thinnest section: two paragraphs, no case evidence.

27.8 Public participation in hazard and options analysis (pp. 659–662)#

• “There are many value judgements involved in hazard and risk analysis, from the framing of the issue and the questions to be addressed to the ethical choice of the appropriate strength of evidence” (p. 659).
• Authoritative bodies recommend public involvement “at all stages of the risk analysis process”: US PCR 1997; RCEP 1998; German Advisory Council on Global Change 2001; Codex 2007; JRC/IPTS 2007; Health Council of the Netherlands 2008; NAS 2009 (p. 659). NAS 2009 (Science and Decisions) “strongly recommends such stakeholder involvement, especially at the crucial problem framing stage” (p. 659).
• Figure 27.2 “A participatory and precautionary framework for analysis of hazards and options” (p. 660). A circle of six segments around a hub labelled “Stakeholder participation and review”:
• [1] framing of hazard questions and control options;
• [2] scientific assessment of hazards, uncertainties and consequences (1+2 = risk assessment);
• [3] option assessment;
• [4] precautionary action (3+4 = risk management);
• [5] communication and implementation;
• [6] effectiveness evaluation of actions and inactions (5+6 = risk communication). Dotted lines show feedback. Source: EEA, based on NRC 1996, US PCR 1997, RCEP 1998 and NAS 2009. The text says engagement intensity is “greater at the problem framing and options choice stages, less so at the scientific risk assessment stage” (p. 659). The figure also puts “consequences” inside the scientific assessment segment, consistent with 27.5.
• Current practice: “These recommendations … do not appear to be reflected in most existing international and European arrangements”, for example for food contaminant limits (JRC/IPTS 2007) (p. 659). European authorities “are continuously improving, albeit at speeds that fail to satisfy all stakeholders”. The Commission consults the public on questions put to risk assessors. Footnote 4 says EFSA runs a “Stakeholder Consultative Platform for food industry stakeholders” and publishes agendas, minutes and opinions (p. 659). The example of stakeholder engagement given here is industry-facing, not public-facing.
• Handling uncertainty. Improvements in food (EFSA 2009; Hart et al. 2010, cited as “Hert”) and emerging issues (SCENIHR 2012), building on the IPCC (pp. 659–660). Floods (Ch 15) show “balancing timely early warnings against false alarms is a very challenging task for decision‑makers” (p. 660). This is the chapter’s main acknowledgement of false alarms.
• Transparency. Improved public access to company data submitted for authorisations (EFSA, January 2013). Openness is “a strong lesson from several case studies — from Minamata to the bio‑ and nanotechnology fields”. The improvements “follow” controversies over aspartame and GM maize (Séralini et al. 2012; Genewatch 2012) (p. 660).
• “From uncertain risks to relevant and responsible innovation?” (p. 660).
• The French GMO debate again.
• Ch 3’s technological “roads not taken” for leaded petrol.
• Ch 19’s “top down” vs “bottom up” agricultural pathways.
• Chs 5 (Minamata) and 16 (bees) question “the value of current democratic institutions in dealing with complex socio‑technical issues”.
• More public involvement “may also lead to wider discussions about technological choices and directions of innovation”.
• “Richard Owens” (Richard Owen), co-author of Ch 13, is developing these ideas in a forthcoming book on responsible innovation (p. 660).
• Box 27.5 (von Schomberg, p. 661):
• The box is “Edited extracts” from von Schomberg (2013). The chapter does not say who edited them.
• Responsible research and innovation (RRI) is “a transparent, interactive process by which societal actors and innovators become mutually responsive”. Social desirability “is currently essentially determined by market mechanisms, however, as universal principles on what counts as socially desirable are not easily agreed upon”. This is a caveat von Schomberg himself states.
• The state is responsible for defining the risks of technologies (through authorisation and liability law), “whereas society lacks a particular responsibility for what could count as positive impacts of technologies”.
• Under the EU “Innovation Union” / Europe 2020, “innovation is assumed to be steerless but inherently good”.
• Societies “lack a specific forum or policy for evaluating particular technologies in terms of benefits and risks”; we have only safety, quality and efficacy hurdles. “The benefits of technologies are ‘demonstrated’ only by market success”, while harms go through formal risk assessment.
• “Modern ‘Frankensteins’ are not intentionally created by a single actor”. Harms arise as “unforeseen side effects of collective action” even with good intentions, so individual-intention or consequence ethics cannot allocate responsibility. What is needed is “ethics of co‑responsibility” for impacts “whether these impacts are intentional or not and whether they can be fully foreseen or not”.
• The challenge is “a more responsive, adaptive and integrated management of the innovation process”. Technical innovators become responsive to societal needs, and “societal actors become jointly responsible for the innovation process”, including defining socially desirable products around challenges “such as climate change and food security”.
• The prescription is a “paradigm shift” in innovation policy. The state “must assume responsibility for positive outcomes of innovation”, reflect public values “beyond consumer market preferences”, and move from technology-oriented policy “towards an issue‑oriented approach” (p. 661).
• Innovation “with a human purpose” (van den Hove 2012) is proposed to rebalance market-focused innovation (p. 662). The public-engagement literature (Wynne, Stirling, Wesselink, Hoppe) “extends well beyond the bounds of the present study”. The Late Lessons evidence “provides further grounds for boosting public and corporate engagement in responsible innovation” (p. 662).

27.9 Conclusion (p. 662)#

• The cases and chapter show “the need for wider use of the PP both as a justification for timely actions on early warnings and as a trigger for broader debates about technological pathways” (p. 662).
• “Mistakes will be made, surprises will occur.” If scientific and stakeholder processes are sound “and the best of science is used”, living with the consequences, “both pleasant and unpleasant, will be more acceptable” (p. 662). This is a procedural-legitimacy argument.
• The capacity to “foresee and forestall disasters appears to be limited”, especially against powerful interests. “It is not just corporations that have the capacity for denial when confronted with evidence of impending disaster — as the financial collapse of 2009 demonstrated” (p. 662). “Wilful blindness” (Heffernan) and “folly” (Tuchman) are general human traits (p. 662).
• “If we adopt optimism of the will to counter pessimism of the intellect … it is possible to believe that human behaviour could improve”. Gramsci’s formula is used without attribution. With humility and engagement, decision-makers could apply the PP more widely and “help anticipate and minimise many future hazards, while stimulating innovation” (p. 662).

References (pp. 662–669)#

• The list is long, and many entries are never cited in the text. A search of the chapter text found no mention of Sunstein 2005 (Laws of Fear), Graham and Hsia 2002, Wiener 2011 (two entries), European Risk Forum 2011, Taleb, Kuhn, Planck, Funtowicz and Ravetz, McMichael, GRADE, Steele, Trouwborst, Zander, Milieu 2011, WTO 2008 (EC–Hormones), several EMF sources (ICNIRP, Adey, Blank, Brauner, Swanson and Kheifets, Neutra, Stewart 2000, Health Council NL 2002, Henshaw), UNESCO 2005, Wingspread (SEHN 1998), Edqvist and Pedersen, or Santillo. Many supportive or neutral works are uncited too, for example Traavik and Li Ching 2007, Umweltbundesamt 2010, Myers et al. 2009 (BPA), RCEP 2008 and Royal Society 2003 (nanotechnology), Aslaksen et al. 2006, Christoforou, Haigh, and Jordan and O’Riordan. So the trimming was general, not aimed only at critics. The published chapter may have been cut down from a longer draft; this is an inference, not stated. Even so, the principal critics of the PP (Sunstein, Graham and Hsia, Wiener, the European Risk Forum) appear only in the bibliography.
• There are minor citation inconsistencies (Kortenkamp 2011/2012; Heffenan 2010 / Heffernan 2011; WBCSD 2010/2011; “Hert” for Hart; Walker 2003 missing).

Case timeline (synthesis chapter: chronology of events and instruments cited, with lags)#

This is not a single case study. The chronology below collects the dated events the chapter uses and the lags it asserts.

Lags the chapter emphasises: - “decades‑long delay between warnings and action” in general (p. 645); - tobacco, more than 40 years (p. 647); - the leaded-petrol research monopoly, about 45–50 years (p. 646); - antibiotic growth promoters, 1965 to 1999 (Table 27.1, p. 656).

Counter-examples of quick action once sentinel human evidence appeared (4–7 cases): DES, VCM, DBCP (p. 645).

The authors’ own lessons and conclusions#

Lessons Gee derives from the case evidence (distinct from advocacy): 1. Late action on known hazards has been very costly in “millions of lives”, environmental damage and economic penalties (p. 644). Asserted here, with the evidence delegated to Ch 23. 2. Common barriers explain “much about the decades‑long delay” between warnings and action: corporate opposition, definitional confusion, complexity, clashing evidence standards, skewed cost analysis, short-termism, lack of public engagement (p. 645). 3. Producers of suspect products have repeatedly run “product defence” campaigns: control of research, loaded language (“natural”, “sound science”), “manufacturing doubt” (pp. 645–646). 4. Where regulators lack independent research, they are vulnerable to corporate influence on the evidence (p. 646). 5. Economic motives dominate corporate responses. Le Menestrel and Rode (Ch 25) found that continuing was “perceived to be profitable” in “virtually all” reviewed cases. Gee adds, in his own voice, that corporate short-term interests prevailed “mainly because” the costs of harm were externalised (p. 647). 6. The historical cases offer only “one or two examples” of responsible corporate behaviour, and those came from companies selling or using hazardous products, not from their manufacturers (p. 647). 7. Early-life exposures cause irreversible, delayed and sometimes multigenerational harm where timing matters more than dose (pp. 649–650). 8. Multicausality undermines the use of absent Bradford Hill features as evidence of no causation. Analogy becomes more important (pp. 652–654). 9. Epidemiology and toxicology methods and funding sources bias results toward false negatives or toward funders’ interests (p. 654). 10. Key harms (mesothelioma, the ozone hole) came from ignorance, not uncertainty (p. 654). 11. Sentinel human cases can drive quick action (p. 645). 12. Early action has stimulated innovation, and late action has entrenched incumbents (p. 659). 13. Lost trust after false reassurance is a major unquantified cost (p. 658). 14. Transparency of regulatory data is “a strong lesson” from Minamata through to biotech and nanotech (p. 660).

Recommendations and advocacy: - Use the PP more widely, in both of its roles (pp. 644, 662). - Adopt the EEA working definition (p. 649). - Reappraise Bradford Hill for complexity, and use “criteria for action” (Box 27.4) alongside criteria for causation (pp. 652–653). - Acknowledge methodological bias in evaluating research (p. 654). - Assess the knowledge-to-ignorance ratio. Where it is low, act on credible early warnings and fund novel research rather than “scientific inertia” (p. 655). - Set case-specific strengths of evidence. Make explicit the purposes and consequences of “not established” judgements. Have risk assessors address consequences and “ramifications” (pp. 656–658). - Widen cost–benefit analysis to unquantifiable pros and cons. Internalise external costs (polluter pays, taxes, charges, permits, assurance bonds) (pp. 647, 658–659). - Separate the political from the economic role of business, and create institutions that expose trade-offs and denial (p. 647). Consider Guidotti’s “room to turn around”, flagged as controversial (p. 646). Encourage shareholder engagement (p. 646). - Create institutions for the long term: ombudsmen and future-oriented committees (p. 659). - Involve stakeholders at all stages of risk analysis, especially framing (Fig. 27.2) (pp. 659–660). Give the public access to company data (p. 660). - Pursue responsible, issue-oriented innovation, with the state taking responsibility for positive outcomes (Box 27.5, von Schomberg) (pp. 661–662). - Shift research funding toward hazard research and alternative pathways such as low-input agro-ecology (p. 646). This is implied by the funding-imbalance critique and by SCAR’s call, which Gee reports. He does not state it as his own recommendation.

Closing position: humility plus engagement plus wider precaution would “anticipate and minimise many future hazards, while stimulating innovation” (p. 662). The chapter frames this as a hope (“optimism of the will”), not a demonstrated result.

Mechanisms and dynamics#

1. How warnings are contested: evidential standards as the battleground. On this reading, the chapter’s central analytical point is that disputes over hazards are often disputes about which standard of proof applies, disguised as disputes about facts. The textual anchor is: “Failing to acknowledge the reality of different strengths of evidence for action has led to several ill‑founded debates” (p. 657). The “disguised” framing is the note-taker’s. - Warnings are dismissed with phrases like “no established or conclusive evidence”, which in practice means the evidence has not convinced a particular committee against an implicit causal-proof standard (pp. 657–658). - The dismissal usually leaves unstated three things: who bears the consequences of being wrong (“risk takers or risk makers”), what the evidence would need to be conclusive for (label, low-cost reduction, ban), and whether the research had even been done (“no evidence of harm” vs “evidence of no harm”) (p. 658). - Table 27.2 turns this into a scale that places legal, trade and scientific standards on one continuum (p. 658). - The Swedish example shows that burden of proof and standard of proof can be split between parties: regulators act on weak evidence of risk, while manufacturers must show safety beyond reasonable doubt (p. 658).

2. Epistemic culture and mental models of experts. The chapter attributes much delay to how scientists and assessors think, not only to interests: - a “monocausal, reductionist” paradigm and “the metaphor of the body as a machine” (pp. 650–651); - statistical practices that strip context and treat co-causes as confounders (p. 651); - misuse of significance testing and neglect of confidence intervals (p. 653); - Bradford Hill’s features used as a checklist for rejection rather than as asymmetric indicators (pp. 652–654); - plausibility tied to “the established knowledge of the day” (p. 653); - “hubris” and lack of “scientific humility” (p. 655).

Together these give a systematic tilt toward false negatives (p. 654). Gee also notes that Hill himself held the more flexible view, “differential standards before we convict” (p. 656), and that followers “have forgotten” his caveats (pp. 653–654). The chapter makes two criticisms at once. The method itself needs reappraisal: Hill’s approach was “essentially based on monocausality” and the criteria “seem less robust now” (p. 652). Its use has also degenerated, because followers forgot Hill’s asymmetry caveats (pp. 653–654).

3. Who produces knowledge, and for what. Several mechanisms shape what is known before any assessment happens: - Control of research by producers. The leaded petrol monopoly left regulators “vulnerable to corporate influence” (p. 646). - Agenda framing toward product development. Only 3 % (EU) or 1 % (USDA) of research spending went on hazards (p. 646). - Property rights over research materials. GM seed access (p. 646). - Disciplinary inertia. A bias toward well-known problems (p. 646) and “excessive research on well known substances” (p. 655). - Funding effects on results (p. 654).

The combined effect is that ignorance is partly produced by how research is organised. Low knowledge about hazards of new products is partly a result of funding choices, not only of novelty.

4. Industry behaviour: product defence and its psychology. - Strategies: control research, adopt reassuring language (“normal”/”natural”, “sound science”), and manufacture doubt from real uncertainty (pp. 645–646). - Motives: perceived profitability of continuing (p. 647, Ch 25); “fear, denial and risk of loss” (p. 646, Guidotti); organisational pressure to deny warnings (p. 647). Economic, epistemological, regulatory, cultural and psychological factors mix (p. 647). - Lock-in to a defensive stance. Once a firm commits to product defence, reversing becomes costlier because of liability exposure. Hence Guidotti’s idea of creating “room … to turn around” (p. 646). This is path dependence at the level of corporate strategy. - Downstream divergence. Companies that sold or used the hazardous product, with less sunk investment in it, moved earlier (asbestos users, CFC aerosols, BPA users) (p. 647). This is a plausible mechanism, but the chapter does not spell it out, and it rests on “one or two” historical examples.

5. Economics: externalisation, asymmetric costs, lock-in. - Harms are externalised to society, and even compensation may be insured (p. 647). - Compliance costs are concentrated, tangible, short-term and fall on “powerful actors”. Benefits are diffuse and long-term (p. 659). This asymmetry biases analysis and political pressure toward inaction. - Unrealistically low prices for hazardous agents (barrier 5, p. 645) plus late action “consolidated technological monopolies” and kept “smarter substitutes” out (p. 659). This is an explicit lock-in / path dependence mechanism linking regulatory delay to suppressed innovation. - Internalisation is presented as the route to earlier substitutes (p. 659).

6. Complexity and causation. The biology itself makes early warnings hard to confirm: - multiple co-causes acting at different stages; - timing-dependent effects (developmental windows); - non-linear and low-dose effects; - bidirectional and circular causation; - cross-scale inference in ecology (pp. 650–651).

Diffuse, multicausal, delayed harms therefore rarely produce the clean signals that triggered quick action in DES, VCM and DBCP: rare “signature” cancers in a handful of people (p. 645; Box 27.2 contrasts tobacco’s single agent with developmental harms, p. 650). The evidential system responds well to specific, rare and prompt signals and poorly to diffuse, common and delayed ones. This is implicit in the chapter’s juxtapositions rather than stated as one thesis.

7. Uncertainty vs ignorance. Some of the worst outcomes were “surprises” that no amount of targeted risk assessment would have found (mesothelioma, the ozone hole) (p. 654). Research expands the frontier of ignorance as well as knowledge (Fig. 27.1, p. 654). The prescribed responses to ignorance are different in kind from reducing uncertainty: - use intrinsic properties (persistence, bioaccumulation, range) as proxies; - use analogies; - monitor over the long term; - prefer “robust, diverse and adaptable technologies” that limit the impact of surprises (Table 27.1, p. 656).

The knowledge-to-ignorance ratio is proposed as the contextual variable that should set how much precaution is appropriate (p. 655).

8. Institutional behaviour and the science–policy boundary. - Jasanoff’s point that “experts”, not “scientists”, govern policy-relevant science (p. 645). - The chapter challenges a strict separation between risk assessment (science) and risk management (values). Assessors should consider consequences and “ramifications” (p. 658), and value judgements run “from the framing of the issue … to the ethical choice of the appropriate strength of evidence” (p. 659). - Regulators’ dependence on industry data and the opacity of that data are treated as institutional failures. Openness is “a strong lesson” (p. 660). - European authorities are credited with incremental improvement: stakeholder consultation, uncertainty guidance, data access (pp. 659–660).

9. Time and short-termism. Electoral and financial horizons are far shorter than hazard latencies and ecological timescales (p. 659). Developmental harms are “often irreversible and sometimes multigenerational” (p. 650). Irreversibility and seriousness are criteria for action (Box 27.4). The combination of long lags, irreversibility and short institutional horizons is a structural driver of late action.

10. Distribution of costs, benefits and risks. - The definition itself requires attention to distribution “across groups, regions, and generations” (p. 649). - Inequitable distribution is a criterion for action (Box 27.4, #9). - Developmental exposure is the extreme case: those harmed (the unborn) get no offsetting benefit (p. 650). - Risk makers vs risk takers is named as the unasked question behind “not established” (p. 658). - Concentrated regulatory costs vs diffuse benefits (p. 659) and externalisation (p. 647) complete the picture.

11. Media and communication. Nuanced two-level findings (suggestive evidence of risk plus no causal interpretation) get collapsed into opposite binary headlines. The chapter blames the scientists’ lack of clarity about evidence levels more than the media (p. 657).

12. Innovation and governance of direction. - The PP’s second role is to open debate on innovation pathways, not only on risks (pp. 644, 660). - Box 27.5 (von Schomberg) adds structural diagnoses: innovation policy treats innovation as “steerless but inherently good”; benefits are validated only by market success while harms face formal assessment; responsibility for collective, unintended harms cannot be assigned under individual-intention ethics (p. 661). - The chapter links research funding priorities to which pathways get developed: agro-ecology vs GM (p. 646; Ch 19), and leaded petrol’s “roads not taken” (p. 660).

13. Framing and language. - The chapter tracks how language does work in hazard disputes: “normal”/”natural”; “sound science” vs “unsound”; “no established evidence”; “no evidence of harm”; “authoritative but unsubstantiated assertions of safety” (Ch 26); “confounders” vs “co-causal factors”; misnamed “criteria”; “triple negatives” in legal definitions (pp. 646, 649, 651–652, 657–658). - Gee’s own framing is also loaded: “vested interests”, “excessively attached to conventional scientific paradigms” (p. 643), “the long history of corporate misconduct” (p. 646), “sin of hubris” (p. 655).

14. Lay knowledge and sentinels. Victims (DBCP) and “an observant clinician” (DES, VCM) supplied the decisive early evidence in the quick-action cases (p. 645). This is a brief but notable acknowledgement that decisive signals often come from outside formal research programmes.

Transferable insights (technology-neutral)#

1. The standard of evidence needed to act should depend on purpose and consequences, not be fixed at the standard of scientific proof. Graduated thresholds (label, low-cost reduction, restriction, ban), set by the costs of being wrong and who bears them, availability of alternatives and policy goals. Evidence: EEA definition (p. 649); Bradford Hill’s “differential standards” (pp. 656–657); legal and trade precedents and Table 27.2 (pp. 656, 658). Strength: strong as a normative and decision-theoretic principle with deep legal and epidemiological precedent. The percentage bands in Table 27.2 are the author’s heuristic, not an established calibration.

2. Burden of proof and standard of proof can be split: weak evidence of risk can license provisional restriction while strong evidence of safety is required of the producer. Evidence: Swedish Chemical Law 1973 at both ends of Table 27.2 (p. 658); WTO SPS Art. 5.7 vs Art. 2 (pp. 649, 658); “provisional” measures in Reg 178/2002 Art. 7 (p. 648). Strength: strong as a description of existing legal designs. Their effects in practice are not assessed here.

3. “No evidence of harm” often means “not studied”, and “not established” statements conceal who bears the risk of error and for what purpose the evidence would be sufficient. Evidence: pp. 657–658; the cross-reference to “authoritative but unsubstantiated assertions of safety” (Ch 26). Strength: strong as a logical and communication point. The empirical frequency across cases is asserted here and evidenced elsewhere.

4. Under multicausality, missing causal indicators (inconsistency, no clean dose-response, low specificity, implausibility under current theory) are weak grounds for dismissal; their presence is informative, their absence much less so. Evidence: pp. 652–654, Box 27.3; the lead variance figure (p. 652); the relative risk 1.5 example (p. 653). Strength: moderate–strong. It fits Hill’s own caveats. But the chapter’s label of Hill’s approach as “essentially based on monocausality” is arguable (the chapter itself qualifies it), and it describes his dose-response feature as a “linear” criterion, which Hill did not require.

5. Conventional study designs and statistical habits skew toward false reassurance. Context stripping treats co-causes as confounders, significance testing converts “no significance” into “no difference”, and plausibility tests privilege existing theory. Evidence: pp. 651, 653–654; reliance on Grandjean (Ch 26). Strength: moderate. The significance-misuse point is well supported. The claim of a systematic net direction toward false negatives is argued, not shown here, and confounder control also guards against false positives, which the chapter does not weigh.

6. Whoever controls, funds or frames research shapes what becomes known, and ignorance about a product’s hazards can be manufactured by funding priorities, not only by novelty. Evidence: the leaded petrol research monopoly (p. 646); 3 % and 1 % hazard shares (p. 646); IP barriers to research materials (p. 646); bias toward known problems (pp. 646, 655). Strength: moderate. The historical case is strong. The 3 % figure is unsourced in this chapter, and the structural generalisation is plausible but not systematically tested.

7. The funding effect: results tend to align with the funder’s interest. Evidence: list of fields and studies (p. 654). Strength: strong for the fields with systematic reviews cited (tobacco, pharmaceuticals, biomedicine). Asserted for cost–benefit analysis and transport (no citation). The chapter itself says the mechanism is unclear (p. 654).

8. Parties with hazardous products at stake have tended to adopt a recognisable defence repertoire: controlling research, reassuring or co-opted language (“natural”, “sound science”), and amplifying uncertainty (“manufacturing doubt”). Evidence: pp. 645–646, citing tobacco, lead, VCM, beryllium, asbestos, benzene and climate cases, Michaels 2008, Oreskes and Conway 2010. Strength: strong for the historical cases (well documented, including through litigation). Suggestive as the chapter’s forward prediction that “other industries … today” will do the same (pp. 645–646).

9. Defensive commitment becomes self-reinforcing through fear, denial and liability, so exit ramps may be needed for actors to change course. Evidence: Guidotti via p. 646; organisational pressure to deny (p. 647). Strength: suggestive. It is one expert’s reflection, and the chapter itself flags the liability-relief idea as controversial.

10. Concentrated, visible, short-term costs of action versus diffuse, long-term benefits (plus externalised harms) systematically bias analysis and politics toward inaction. Evidence: pp. 647, 658–659; Stern via Ch 14. Strength: strong as political economy, consistent with a large literature. The chapter illustrates rather than measures it.

11. Delay entrenches incumbents: late action lets a hazardous option consolidate at prices that exclude its hidden costs, keeping better substitutes out, while earlier action can pull substitutes forward. Evidence: p. 659 (asbestos, lead, mercury, PCBs, CFCs, benzene); Porter and Ashford literature. Strength: moderate / suggestive. The cases are cited without data here, and the general “regulation spurs innovation” claim is contested in its strong form.

12. Ignorance differs in kind from uncertainty. Some of the gravest harms were surprises, and research expands the frontier of ignorance as well as knowledge. Responses to ignorance must rely on proxies (persistence, mobility, accumulation, scale), analogy, monitoring, and diverse, adaptable, reversible options, not only on better risk estimates. Evidence: p. 654, Fig. 27.1; Table 27.1 (p. 656). Strength: strong conceptually (well-chosen historical surprises). Moderate as operational guidance, because it is not specified how to apply it.

13. How much precaution is appropriate depends on how much is known relative to what is unknown, and on whether credible early warnings exist. Evidence: pp. 643, 655; Box 27.4 #2. Strength: suggestive. The KIR is unquantifiable by the author’s own admission. Using “novelty” as a criterion risks treating newness itself as hazard, although the chapter conditions action on “credible early warnings” and “plausible evidence” (p. 655).

14. Intrinsic, observable properties and analogies with known harms can serve as early-action triggers when causal proof is unavailable. Evidence: Box 27.4 (p. 653); Table 27.1 (p. 656). Strength: moderate. Persistence, bioaccumulation and range criteria have a track record in chemicals governance. The twelve-item list as a whole is unweighted and untested, and it mixes hazard evidence with policy considerations (innovation potential, alternatives).

15. Decisive early signals often come from specific, rare, prompt outcomes seen by frontline observers or affected people. Diffuse, common, delayed, multicausal harms rarely produce such signals and are structurally disadvantaged by evidence rules. Evidence: p. 645 (DES, VCM, DBCP: 4–7 cases); Box 27.2 (p. 650); pp. 650–651. Strength: moderate. The quick-action cases are well documented. The contrast is an inference from the chapter’s material rather than its stated thesis.

16. False reassurance carries a large cost: loss of trust in expert bodies after harm they said would not occur. Evidence: p. 658 (BSE, Chernobyl, Fukushima). Strength: moderate. BSE is a strong exemplar. The comparative magnitude (“as significant as the economic costs”) is asserted.

17. Value judgements permeate risk analysis from framing onward, so framing and choice of evidential standard are legitimate sites for broader participation. Assessors should convey the consequences of their judgements, not only causal verdicts. Evidence: pp. 658–660, Fig. 27.2; NAS 2009 and other bodies; European Court “ramifications” (p. 658). Strength: moderate. It is backed by authoritative reports. Evidence that participation improves outcomes is not presented, and the chapter notes practice lags (p. 659).

18. Institutional time horizons (electoral, financial) are mismatched with hazard latencies and irreversibilities. Evidence: p. 659; latency and irreversibility material (pp. 649–650); the decades-long lags across the cases (pp. 645–647). Strength: moderate–strong as diagnosis. The mismatch is plausible given the documented lags, but the chapter asserts it in two paragraphs and does not show how short horizons caused particular delays. Suggestive for the remedies (future-generation ombudsmen), which are only listed.

19. Innovation systems validate benefits through market success while subjecting harms to formal assessment, and responsibility for unintended collective effects falls between actors. Evidence: Box 27.5 (von Schomberg, p. 661). Strength: moderate as a structural observation. Asserted/normative for the prescription (state responsibility for positive outcomes, issue-oriented innovation policy).

20. Communicating graded evidence as a binary invites polarised misreading. Findings should state explicitly which level of evidence they reach and for what purpose. Evidence: Interphone and the headlines (p. 657). Strength: suggestive. It rests on one illustrative case, quoted selectively (see below).

21. Downstream users of a hazardous input, with less sunk commitment, often move before producers, which makes supply-chain actors a potential lever for earlier change. Evidence: p. 647 (asbestos users, aerosols, BPA users). Strength: suggestive. The examples are anecdotal: the chapter itself says the historical cases show only “one or two examples”. At least one is probably misattributed, and the more recent examples (BASF, Hewlett Packard) include manufacturers.

Limitations, contestation and bias check#

Advocacy vs analysis. - The chapter is explicitly a case for more precaution by the project’s originator and editor. Several of the case chapters it summarises were co-written by the same author (TOC, PDF pp. 5–6: Chs 3, 8, 21). Other members of the editorial team supply key supporting arguments: Grandjean (Ch 26) on false negatives, MacGarvin on “hubris”, van den Hove on innovation “with a human purpose”. It is an editorial synthesis, not an independent evaluation. - The summary’s characterisation of opponents (vested interests; scientists “excessively attached to conventional scientific paradigms”, p. 643) explains opposition by interest or intellectual failure. It does not consider principled objections: risk–risk trade-offs, opportunity costs, legal certainty, trade protectionism, regulatory inconsistency. - The title poses “more or less?” but the text never argues the “less” side. The best-known critics (Sunstein 2005; Graham and Hsia 2002; Wiener 2011, whose “real pattern of precaution” argues that jurisdictions are selectively rather than generally precautionary; the European Risk Forum 2011) are in the reference list but absent from the text. The debate the title invokes is signalled bibliographically but not engaged.

Balance the chapter does show (fairness in the other direction): - It concedes Carson “overstates the situation” (p. 645). - It hedges one of its own precautionary examples: the antibiotics ban is only “arguably” precautionary, and “belated” (p. 644). - It flags Guidotti’s liability-relief idea as likely to “be controversial” (p. 646). - It reports Le Menestrel and Rode’s warning against hindsight blame (p. 647). - It lists responsible corporate actions (p. 647). - It acknowledges European authorities are improving (p. 659). - It notes that balancing early warnings against false alarms is “very challenging” (p. 660). - It concedes “Mistakes will be made, surprises will occur” (p. 662). - It widens denial beyond corporations (p. 662). - It sets a floor: “All responsible applications … require some plausible evidence” (p. 655). - It insists precaution is not needed once causation is proven, keeping precaution and prevention distinct (p. 657).

Thin or unsourced evidence. - The “3 % of EUR 28.5 billion” figure has no citation (p. 646). - “Millions of lives” and “very large economic penalties” are asserted, with the evidence delegated to Ch 23 (p. 644). - The “88 claimed false positives” are said to show “the value of the PP”, but the finding is not stated or explained here (p. 645). - The innovation-stimulus and monopoly claims cite cases and literature but no data (p. 659). - Funding bias in cost–benefit analysis and transport has no citation (p. 654). - “Scientists in many of the case studies” guilty of hubris: no cases named (p. 655). - The EEA definition “has proved useful”: no evidence of uptake (p. 649). - Short-termism remedies are listed, not evaluated (p. 659). - The “common myths” promised in the summary (p. 643) get no section of their own. Only the North Sea misreading is rebutted explicitly (p. 657).

Conceptual issues. - Table 27.2 puts heterogeneous things on one probability scale: legal standards of proof, IPCC-style likelihood language, insurance, and “statistical significance” labelled “very strong (90–99 %)” (p. 658). Treating a significance level as the probability that a causal claim is true is itself a common misreading of p-values. That is awkward in a chapter criticising misuse of significance (p. 653). - “Balance of probabilities” (by definition above 50 %) sits in a 33–65 % band. - The table mixes evidence of risk and evidence of safety in one column. That is illuminating for burden of proof but confusing as a scale. - The chapter says the bands are “based on the IPPC [sic] approach” (p. 657), and they match the IPCC 2001 likelihood bands. But the IPCC separates “likelihood” from “confidence” (evidence and agreement), and the chapter merges them as “strength of evidence”. - Bradford Hill. Calling his dose-response feature a “linear” criterion (p. 652) is not how Hill put it. From background knowledge of Hill (1965): he wrote of a “biological gradient, or dose-response curve”, illustrated it with the roughly linear rise of lung-cancer deaths with cigarettes smoked, and did not make linearity a requirement. Describing his approach as “essentially based on monocausality” is arguable. The chapter itself immediately says Hill “was aware that several factors would be implicated” and quotes him on multiple causes (p. 652). Gee’s central point, that absence of features is not evidence against causation, matches Hill’s own caveat that no single viewpoint is indispensable. - Confounding. Reframing confounders as “co-causal factors” (p. 651) is a real issue in complex exposures. But confounder control also prevents spurious associations (false positives), and the chapter does not weigh that trade-off. - Box 27.4 mixes hazard indicators (toxicity, persistence) with policy considerations (“potential for stimulating innovation”, “availability of feasible alternatives”). These are legitimate inputs to a decision but not features of evidence about hazard, as the box title claims. Including “novelty” as a criterion invites the objection that precaution becomes a presumption against the new. The chapter’s answer (act “following credible early warnings”, p. 655) is not built into the box. - Not addressed: when and how precautionary measures should be lifted or revised as evidence accumulates. Only “provisional” in Box 27.1 and “time scales for future learning” (Box 27.4 #12) gesture at it. Risk–risk trade-offs and “regrettable substitution” (replacing one hazardous agent with a less-studied analogue) are not discussed either. Formally, the EEA definition’s “pros and cons of action and inaction” (p. 649) and “availability of feasible alternatives” (Box 27.4 #10; p. 656) leave room for them. But the chapter gives no example where a precautionary action itself caused harm or cost more than it saved.

Hindsight and case selection. - The lessons draw on cases chosen because harm materialised. The chapter points to the false-positives analysis (p. 645) as a counterweight but does not engage it. - Two of the five “precautionary actions” are contested as successes: - the EU hormones ban was found by the WTO Appellate Body (1998) not to be based on a risk assessment (the 2008 AB report is in the reference list but not discussed); - EU GMO restrictions are presented as illustrating “the value of the PP in minimising harm”, although the evidence of averted harm is disputed. - The chapter’s own KIR examples of emerging hazards (GM food, mobile phones, nanotechnology) are presented as a “vast ocean of ignorance” (p. 655). Later evidence has not uniformly borne out serious harm (see below). The framework’s value may lie more in its logic than in these specific calls.

Selective quotation (Interphone, p. 657). - The published abstract’s conclusion (checked via Europe PMC, PMID 20483835) begins: “Overall, no increase in risk of glioma or meningioma was observed with use of mobile phones”. It continues that there were suggestions of increased glioma risk at the highest exposure levels, “but biases and error prevent a causal interpretation”. - The chapter’s version omits the “overall, no increase” sentence. It uses a different wording, possibly from the paper’s main-text conclusions, with an ellipsis (“….”) that drops the exposure qualifiers. - The BBC headline therefore tracked the study’s own lead sentence, and the Telegraph the “suggestions” sentence. This can be read two ways. It is consistent with Gee’s point that a two-level finding gets read as a binary. But the truncated quotation makes the study look more suggestive of risk than its headline conclusion was. It also lets the chapter blame the scientists’ lack of clarity without showing that the study itself led with “no increase”. The quotation is selective; the underlying communication point survives.

Factual errors or doubtful details (verify before reuse): - “Johnson & Johnson stopped using CFCs in their aerosols in 1977” (p. 647). The well-known early voluntary phase-out was by S.C. Johnson & Son (Johnson Wax) in 1975, in response to the 1974 Rowland–Molina warning (checked in a secondary source, Wikipedia’s S. C. Johnson Jr. entry). This is probably a confusion of companies and date. Cross-check against the LL1 halocarbons chapter. - “before the European Commission took action on its use in baby toys” (p. 647). The 2011 Commission action on BPA was Directive 2011/8/EU restricting BPA in polycarbonate infant feeding bottles (checked on EUR-Lex). “Toys” is probably an error for “bottles”. - TBT ban in France “in 1984” (p. 644). France’s restriction on TBT paints for small boats is usually dated 1982. Verify against the LL1 TBT chapter; not re-verified here. - IMO Anti-fouling Systems Convention “2000” (p. 648). The convention was adopted in 2001. From background knowledge; minor. - “financial collapse of 2009” (p. 662): conventionally 2007–08. - Israel’s Commission for Future Generations (cited as a current experiment, p. 659) operated roughly 2001–2006 and had lapsed before 2013. From background knowledge; verify. - Blue vs white asbestos (fn 2, p. 655): the claim that the potency difference “practically disappears” (Lenters et al. 2012) is one side of a continuing epidemiological dispute, not a settled finding. - Séralini et al. 2012, cited as a controversy prompting transparency (p. 660): the underlying rat-feeding paper was retracted by Food and Chemical Toxicology in late 2013 and later republished elsewhere. From background knowledge; the chapter does not endorse the study’s findings. - “ethinyl oestradiol in the pregnancy pill” (p. 644): the chapter itself says “contraceptive pill” at p. 649. - “hormones in cattle feed” (p. 644). From background knowledge, the EU ban concerned hormonal growth promoters, which were mainly given as implants rather than in feed. Minor imprecision. - “ECR, 1999 and 2002” (p. 658). The reference list shows both are Court of First Instance cases (T‑13/99 Pfizer, T‑70/99 Alpharma). From background knowledge, both judgments date from 2002. - Name slips: “Richard Owens” for Richard Owen; “Baruoki” for Barouki; “Heffenan”; “Rene Schomberg”; “Constanza” for Costanza; “Genewatch 2012” in the text but “GM Watch, 2012” in the references.

Later evidence (post-2013; outside the report). Items marked [checked] were verified in this session. The others are from background knowledge and should be verified before use. - BPA (named as an emerging risk, p. 644). - [checked] EFSA’s 2023 re-evaluation set a TDI of 0.2 ng/kg bw/day, “20 000 times lower” than its 2015 value, and found dietary exposure exceeds it by two to three orders of magnitude. - [checked] Commission Regulation (EU) 2024/3190 (19 December 2024) broadly bans BPA and other hazardous bisphenols in food contact materials. - This strongly supports the chapter’s direction on BPA. Scientific disagreement persisted: other agencies, reportedly including BfR and EMA, differed with EFSA’s 2023 TDI (verify). The regulation’s extension to other bisphenols also shows awareness of regrettable substitution. - Neonicotinoids (p. 644): EU partial restrictions in 2013 and an outdoor ban on imidacloprid, clothianidin and thiamethoxam in 2018. Consistent with the chapter’s call. - Endocrine disrupters (p. 644): EU identification criteria for biocides and plant protection products (2017–18); new CLP hazard classes for endocrine disruption (2023). Consistent with the chapter’s direction. - Mobile phones (pp. 644, 655, 657): IARC classified RF-EMF as “possibly carcinogenic” (2B) in 2011, before publication and not mentioned here. A WHO-commissioned systematic review (Karipidis et al., 2024) found no association between mobile phone use and brain cancers. This cuts against the chapter’s implied emphasis on possible risk, but not against its point about communicating graded evidence. - GM food (p. 655): EU-funded two-year rat feeding studies (e.g., G-TwYST) reported no adverse effects of the tested maize. Major scientific bodies continue to judge approved GM foods as safe as conventional counterparts. This weakens the chapter’s treatment of EU GMO restrictions as a precautionary success on health grounds, though not its point about debates over agricultural pathways. - Nanomaterials (p. 655): some specific precautionary or regulatory actions followed. For example, EFSA (2021) concluded titanium dioxide (E171) could no longer be considered safe as a food additive, and the EU banned it in 2022. This is partial support for the low-KIR argument, applied material by material rather than to the technology class. - Antibiotic growth promoters (pp. 644, 656): the full EU ban took effect in 2006. International bodies later recommended against growth-promotion use. This strongly supports this precautionary case. - Statistical significance (p. 653): the American Statistical Association’s 2016 statement on p-values and the 2019 call to “retire statistical significance” (Amrhein, Greenland, McShane, Nature) strongly vindicate the chapter’s critique. - Bradford Hill reappraisal (p. 652): later literature (e.g., Fedak et al. 2015) revisited the viewpoints for 21st-century data integration, broadly in the direction Gee urged. - IPCC (p. 657): AR5 (2013) found it “extremely likely” that human influence was the dominant cause of recent warming, and AR6 (2021) called it “unequivocal”. This confirms the chapter’s example of graded evidence strengthening over time. - Funding effect (p. 654): a Cochrane review (Lundh et al. 2017) confirmed that industry-sponsored drug and device studies more often report favourable results. This supports the chapter for pharmaceuticals. - Porter hypothesis (p. 659): later reviews found the “weak” version (regulation stimulates innovation) reasonably supported and the “strong” version (net competitiveness gains) mixed. This qualifies the chapter’s innovation claim. - Transparency (p. 660): the EU Transparency Regulation (EU) 2019/1381 strengthened public access to studies submitted to EFSA, confirming the direction the chapter noted. - Participation and RRI (pp. 659–662): RRI became a cross-cutting theme in Horizon 2020 (2014–20) but was less prominent as a named programme in Horizon Europe. Meanwhile an industry-promoted “innovation principle” gained a foothold in EU policy language (the European Risk Forum, in this chapter’s bibliography, was a promoter). The policy contest over precaution continued rather than resolving in the chapter’s favour. - Long-term institutions (p. 659): the Well-being of Future Generations (Wales) Act 2015 and its commissioner, and the UN Declaration on Future Generations (2024), show continued institutional experimentation.

Notable quotes#

1. “the application of the precautionary principle has been strongly opposed by vested interests who perceive short term economic costs from its use” (p. 643)
2. “Taken together, these barriers explain much about the decades‑long delay between warnings and action.” (p. 645)
3. “Without access to independent research the regulatory authorities were vulnerable to corporate influence on the scientific evidence made available to them.” (p. 646)
4. “the way in which technological and hazard problems are framed can result in research that focuses much more on developing products than on the need to find out whether those products are harmful” (p. 646)
5. “The presence of the criteria can be robust evidence for a causal association, whereas the absence of the criteria is not robust evidence that there is no causal association.” (p. 653)
6. “Learning to live with and manage irreducible uncertainties is as necessary as trying to reduce them.” (p. 654)
7. “The different consequences for those for whom the evidence is ‘not established’ (i.e. risk takers or risk makers) is seldom discussed.” (pp. 657–658)
8. “Once evidence reaches the level of ‘proving’ a causal link there is no need for the PP as the issue is then firmly in the ‘prevention principle’ area” (p. 657)
9. “late actions have consolidated technological monopolies for products, at unrealistically low prices, which served to keep smarter substitutes out of the markets for many years” (p. 659)
10. “The benefits of technologies are ‘demonstrated’ only by market success, whereas the potential negative consequences are evaluated under formal risk assessment schemes.” (Box 27.5, von Schomberg, p. 661)

Open questions#

1. How should the “appropriate strength of evidence” be chosen in practice, and by whom? The chapter gives inputs (pp. 656–657) and a participatory frame (Fig. 27.2) but no procedure. Did later EU practice (EFSA uncertainty guidance, SCENIHR weight-of-evidence) operationalise this?
2. What does the “88 claimed false positives” analysis (Ch 2) actually show, and does it answer the selection-bias objection as the chapter implies (p. 645)? Cross-check with the LL2-02 notes.
3. Where does the “3 % of EUR 28.5 billion” figure come from (p. 646), and what has the hazard-research share of public funding been since 2013?
4. How can the knowledge-to-ignorance ratio be assessed consistently so that “novelty” does not become a blanket presumption against new technologies (pp. 655, Box 27.4)? Were any assessments ever done on this basis?
5. How should precautionary measures be reviewed and lifted when evidence later points to low risk (e.g., the mobile phone and GM food examples)? The chapter is silent on exit criteria and on regrettable substitution.
6. Do participatory framing processes measurably change which hazards are acted on, or how quickly? The chapter relies on authoritative recommendations, not outcome evidence (pp. 659–660).
7. Is the downstream-users-move-first pattern (p. 647) real and general? Why do users move before manufacturers?
8. Guidotti’s “room to turn around” (p. 646): are there documented cases where liability relief or safe harbours induced earlier corporate reversal, and at what cost to victims’ justice (compare Ch 24)?
9. Table 27.2: is a single probability-like scale for “strength of evidence” coherent across legal, statistical and scientific-assessment standards, or should likelihood and confidence be kept separate, as the IPCC does?
10. Two of the chapter’s five “precautionary actions” are contested as successes (hormones, GMOs): what would a fair, symmetric scorecard of precautionary actions, including costs, look like?
11. How did the “innovation principle” versus precautionary principle contest in EU policy after 2013 play out, and does it bear out the chapter’s barrier 1 (corporate opposition) or show principled disagreement?

Audit log#

Independent audit against the full text extract (PDF pp. 645–671). Also checked: the PDF acknowledgements (p. 7), TOC (pp. 5–6), Annex 1 (p. 691), Figure 27.1, Table 27.1, Table 27.2, and the Interphone abstract (Europe PMC, PMID 20483835). All quotations in “Notable quotes” and the section notes were checked verbatim. Reference-list claims were checked by searching the chapter text.

• Authors: replaced the paraphrase “originating idea” with the verbatim acknowledgements line (PDF p. 7); added that Grandjean, MacGarvin and van den Hove, whose work the chapter leans on, sat on the editorial team.
• Summary box: fixed a malformed quotation (“criteria for action”).
• Summary box: softened “never delivers … myths” / “only misreading”: no dedicated section, but the chapter does address misreadings piecemeal (pp. 654, 657, 658).
• 27.1 case recap: corrected a garbled sentence. It is the known-hazard cases that “primarily illustrate how more precautionary action could be applied to chemical risks emerging now”. Flagged “pregnancy pill” vs “contraceptive pill” (p. 649).
• 27.1: added the omitted sentence on alien species, floods, ecosystems and climate change (p. 644).
• 27.1: renamed “Precautionary successes” to “Precautionary actions”. The chapter says the five cases “describe precautionary actions”, not “successes”.
• 27.1: added that the DES/VCM/DBCP actions came “only after serious and compelling human evidence”, i.e. fast prevention, not precaution.
• 27.1 barrier 7: restored the qualifier “a failure in most cases”.
• 27.2 responsible behaviour: added the chapter’s “one or two examples” qualifier, and noted that the “not manufacturers” caveat applies to the historical examples only; several recent ones are manufacturers.
• 27.4: restored the hedge “appear to involve some or all of” for the eight disease-process events.
• 27.4 Bradford Hill: added the chapter’s own qualification (Hill “was aware that several factors would be implicated”) and its line that the criteria “seem less robust now” than in the 1960s.
• 27.5 Interphone: added “13-country”; attributed the “similar confusion” framing to Gee.
• 27.5 IPCC: added the chapter’s statement that Table 27.2’s five bands are “based on the IPPC [sic] approach”.
• 27.5 European Court: added Gee’s reply to the risk-manager objection; identified “ECR 1999 and 2002” from the reference list as CFI cases T‑13/99 and T‑70/99; flagged that both judgments date from 2002 (background knowledge).
• 27.8: noted that fn 4 describes EFSA’s platform as “for food industry stakeholders”.
• Box 27.5: added “edited extracts”; von Schomberg’s own caveat that social desirability is hard to agree; the state/society asymmetry on risks vs positive impacts; and the “jointly responsible” / beyond-market-preferences wording.
• References: added that many supportive or neutral works are also uncited, so the trimming was general, not aimed only at critics. Marked the “cut down from a longer draft” point as an inference.
• Timeline: corrected the “Pfizer case” 1999 entry (case lodged 1999; judgment 2002).
• Lessons #5: separated Le Menestrel and Rode’s finding (perceived profitability) from Gee’s own causal claim (externalised costs); the draft had conflated them.
• Lessons #6: replaced the overstatement “downstream users provide most examples” with the chapter’s “one or two examples”.
• Recommendations: marked the research-funding shift as implied, not stated as Gee’s own recommendation.
• Mechanism 1: anchored the “standard-of-proof disputes” reading to the chapter’s “ill‑founded debates” sentence (p. 657), and marked “disguised” as the note-taker’s framing.
• Mechanism 2: corrected a misreading. The chapter criticises both Hill’s method (needs reappraisal) and its degenerate use, not only the latter.
• Mechanism 4: replaced “retailer” (not in source) with sellers/users; noted the thin base of “one or two” historical examples.
• Insight 4: softened “misdescribes”; the chapter qualifies its own monocausality claim.
• Insight 18: downgraded “strong” to “moderate–strong”, because Section 27.7 is two paragraphs with no case analysis.
• Insight 21: added the “one or two examples” qualifier and the manufacturer counter-examples.
• Limitations, balance: added the “arguably” hedge on the antibiotics ban and the “controversial” flag on Guidotti’s idea.
• Limitations, advocacy: added the editorial-team overlap and named the co-authored chapters (3, 8, 21).
• Limitations, thin evidence: reworded the “myths” point.
• Limitations, Table 27.2: replaced “appear to follow the IPCC” with the chapter’s explicit statement.
• Limitations, Hill: added that Hill used a linear example but did not require linearity (background knowledge).
• Limitations, not addressed: noted that the definition’s “pros and cons of action” and “feasible alternatives” formally leave room for risk–risk trade-offs, but the chapter gives no example of precaution causing harm.
• Limitations, Interphone: rebalanced the critique. The quotation is selective, but the binary-misreading point survives; removed the claim that this simply “weakens” the case.
• Errors list: added “pregnancy pill”, “hormones in cattle feed” (growth promoters were mainly implants; background knowledge), the ECR dating, “Constanza”, and Genewatch/GM Watch.
• Hindsight section and open question 10: replaced “precautionary successes” wording with the chapter’s “precautionary actions”.
• Digest: header now says “edited extracts” and names the three co-authored chapters.
• Digest, key evidence: separated the Ch 25 finding from Gee’s externalisation claim; “infertility” changed to the source’s “sperm reduction”; lost trust now uses the source’s “can sometimes be as significant as” rather than “major”; the “hidden choices” wording is now faithful to p. 658.
• Digest, mechanisms: fixed the “risk takers rather than risk makers” misstatement (the chapter says their differing consequences are “seldom discussed”); anchored the standard-of-proof mechanism to p. 657.
• Digest, insights: qualified #3 and #7; flagged #10 as an inference; downgraded #13 to moderate–strong.
• Digest, caveats: added a “some balance shown” line; changed “nothing on” to “little on”; renamed “Contested successes” to “Contested examples”; added the “balance of probabilities” band anomaly, the Hill “linear” point, the TBT and 2009 date issues, and a more nuanced Interphone line.
• Strand check: no mentions of off-limits contemporary technologies or companies found. “Information technology”, Hewlett Packard and the EFSA “platform” come from the source.