Science Follows the Buyer

TL;DR

The problem is not that industry fails to do science. Firms do enormous amounts of it, including safety research when market access requires evidence. The structural gap appears when the relevant question concerns delayed, cumulative, population-wide, or ecological effects that no individual company can easily measure, capture, or profit from. Product safety and system consequences are different research objects. Sponsorship can bias evidence, but the deeper problem is agenda ownership: optimization has a customer; diffuse consequence science often does not. Public research, pre-market requirements, post-market surveillance, and producer-responsibility rules are ways of creating one.

Laboratory precision is real, but precision within a narrow question does not guarantee that the larger system asked the right one. Photo: Polina Tankilevitch on Pexels.¹

Industry Does Science Where Decisions Live

A food company trying to extend a yogurt’s shelf life from 14 days to 30 has a tractable research problem. It can vary temperature, cultures, packaging, stabilizers, sanitation, and transport conditions. The result appears quickly in spoilage rates, returns, distribution radius, and margin.

The customer for the knowledge is obvious. The question has an owner, a budget, and a decision attached to it.

Now ask whether a combination of processing, packaging, product substitution, and eating patterns contributes to a modest rise in metabolic disease over 25 years. The effect, if it exists, must be separated from income, exercise, smoking, medication, genetics, total diet, and changing diagnosis. The costs of the study are concentrated. The benefits of knowing are spread across competitors, consumers, insurers, and governments. A company that pays for the answer may discover a liability or hand useful knowledge to rivals.

This is not evidence of uniquely corporate wickedness. It is an incentive problem.

Economists describe part of it as a spillover. Research produces knowledge others can use, so the investing firm cannot capture the full social return. The OECD’s review of R&D policy notes that firms typically underinvest when benefits spill to other firms, costs are high, returns are uncertain, and financing is difficult.² Consequence research adds a harsher version: the result may reduce the value of the very product paying for the study.

Science follows the buyer because science is labor, equipment, data, time, and institutional permission—not disembodied curiosity.

Consumer Choice Is Not a Neutral Laboratory

The usual market reply is that companies may optimize whatever they like and consumers can choose among the results. That answer works better for some goods than others.

Consider breakfast cereal. The difficult engineering questions include extrusion, coating, texture after contact with milk, color stability, flavor release, packaging, shelf life, and manufacturing consistency. The commercial questions include mascot design, placement, brand memory, portioning, and repeat purchase. A company can measure many of those outcomes within weeks.

The socially important question is different: How do we make something a child can eat regularly for years with the best plausible health outcome? That research program might optimize intact-grain content, fiber diversity, satiety, protein quality, added sugar, and long-term dietary substitution. Some firms do pursue those goals. But their returns are harder to observe and capture than sales, margin, or shelf stability.

Calling the final purchase a free choice hides how the preferences entering that choice were produced. One side can employ sensory science, package testing, advertising experiments, retail placement, and years of brand conditioning. On the other side may be a six-year-old and a tired parent moving through a supermarket.

This is not a speculative objection to advertising. The World Health Organization’s evidence-based guideline reports that food marketing affects children’s food choices and dietary intake, their purchase requests to adults, and the norms they develop about food consumption. It recommends policy protection rather than treating children as miniature informed consumers.³

None of this makes the consumer powerless or every appealing product manipulative. It means choice is an outcome of a system as well as an input to it. When companies can scientifically shape demand, “the market merely gives people what they want” becomes circular: market actors helped cultivate what people came to want.

Safety Science Is Not Fake Science

Critiques of industry often become careless here. Modern food and agricultural systems contain extensive testing. Pasteurization, sanitation, refrigeration, toxicology, residue limits, packaging migration studies, and pathogen controls prevent real harm. A processing intervention that extends shelf life can reduce food waste and foodborne disease. “Industrial” is not a synonym for dangerous.

Nor is company-generated evidence automatically invalid. Regulators deliberately require firms to generate it. The US Environmental Protection Agency, for example, requires pesticide data covering repeated and chronic exposure, carcinogenicity, non-target organisms, environmental exposure, and applicator risk. Some studies progress from laboratory tiers to field conditions when predicted harm warrants it.⁴

The US Food and Drug Administration similarly reviews laboratory, animal, and human testing performed by companies for new drugs and several other regulated product classes. It also makes clear that regulatory regimes differ: some products require pre-market proof, others must meet performance standards, and some can enter the market without prior approval.⁵

That variation is central. “Has it been studied?” is too vague. Better questions are:

Studied for which endpoint?
Over what duration and exposure range?
As an isolated component or as part of a lived system?
Who selected the comparison and outcome?
What happens after adoption reaches millions of people?

A rigorous answer to a narrow question remains rigorous. It is not an answer to every adjacent question.

Products Are Tested One at a Time; People Live in Systems

Regulation often evaluates a product, compound, or process against specified hazards. This is necessary because decisions need bounded objects. But human and ecological experience is cumulative.

A person encounters a diet, not one additive; a watershed receives runoff from a region, not one farm; antibiotic resistance emerges across hospitals, farms, microbes, prescriptions, and trade. A facility can satisfy its individual limits while the combined system produces an effect no single facility controls.

The difference is not merely “short term” versus “long term.” It is product causation versus system causation.

Product causation is often experimentally accessible: does this dose produce this outcome under these conditions? System causation involves feedback, substitution, behavior, heterogeneous exposure, and adaptation. A longer laboratory study does not automatically solve it. Sometimes the thing needing study did not exist until millions of users reorganized their lives around it.

This is why pharmaceuticals pair pre-market trials with post-market surveillance. FDA notes that approval studies usually involve relatively small groups and that some problems emerge only after large-scale use. Companies therefore must submit post-market safety information.⁶ The institutional design concedes a basic fact: adoption creates evidence that cannot exist before adoption.

Food systems, digital platforms, chemicals, and urban technologies vary greatly in how completely they build that learning loop.

Ultra-Processed Food Shows Both the Discovery and the Limit

The debate over ultra-processed food illustrates why neither complacency nor panic is warranted.

In a controlled NIH inpatient trial, 20 adults received an ultra-processed diet for two weeks and an unprocessed diet for two weeks, in randomized order. The offered diets were matched on several presented nutritional measures. Participants consumed about 508 more kilocalories per day during the ultra-processed phase and gained about 0.9 kilograms, while losing about the same amount during the unprocessed phase.⁷

This is unusually strong evidence for a specific causal claim: under those study conditions, the ultra-processed menu changed how much participants ate and their short-term weight trajectory.

It does not prove that every product classified as ultra-processed has the same effect. It does not isolate a single mechanism. It does not establish a 25-year disease risk, compare every cuisine, or show that processing as such is harmful. Twenty people over four weeks cannot carry those conclusions.

The study is valuable precisely because it narrows uncertainty without pretending to abolish it. It also demonstrates the agenda problem. Food science had long optimized texture, preservation, palatability, consistency, and cost. Asking whether a whole menu’s processing structure alters spontaneous intake required a public research institution to assemble a residential experiment around a social consequence rather than a product improvement.

The research did not reveal that earlier science was fraudulent. It asked a question the earlier optimization objective did not contain.

Sponsorship Changes More Than the Result

Money can also influence what evidence says.

A Cochrane review of 75 papers found that industry-sponsored drug and device studies more often reported favorable efficacy results and favorable conclusions than non-industry-sponsored studies. Conventional risk-of-bias measures did not explain the difference, and evidence about harms was less precise.⁸

That finding does not justify discarding every company-funded study. Sponsorship can correlate with product selection, dose, comparator, outcome choice, publication, and interpretation as well as laboratory quality. A firm may reasonably fund the candidate most likely to succeed. It may also compare against a weak alternative or emphasize the endpoint most useful for approval.

The research agenda is therefore part of research bias.

If ten precise studies examine whether a new package preserves crispness and none asks how its waste accumulates, the missing evidence will not appear as a false result. It appears as an unasked question. Peer review can improve a paper that exists; it cannot review a study nobody commissioned.

Regulation Can Create a Buyer

The incentive gap is not destiny. Institutions can attach consequence knowledge to a decision or liability.

Pre-market evidence requirements make market access conditional on answering specified safety questions. They work best when hazards are foreseeable, tests are informative, and delaying entry is justified by potential harm.

Post-market surveillance creates a continuing duty to learn from large-scale exposure. It is essential when rare or delayed effects cannot be resolved in small pre-deployment studies.

Independent public research pays for questions whose benefits are broadly shared or whose answers may be commercially inconvenient. Public funding need not replace private R&D; it can direct attention toward spillovers that markets undervalue. OECD regional evidence even suggests that government R&D can stimulate additional private investment rather than simply displace it.²

Extended producer responsibility changes the objective by assigning producers responsibility beyond the point of sale. The OECD defines it as responsibility across the product lifecycle, including the post-consumer stage.⁹ When disposal becomes a producer cost, research on durability, repair, recovery, and material choice acquires an internal customer.

Shared levies and research pools can prevent a conscientious firm from paying alone for knowledge that benefits an entire sector. Governance must protect the agenda from capture, but collective funding can match collective exposure.

Each mechanism has failure modes. Pre-market rules can freeze useful innovation or reward firms large enough to afford compliance. Surveillance can drown in noisy reports. Public research can follow political fashion. Producer-responsibility schemes can become accounting exercises. The answer is institutional scrutiny, not the fantasy of a funding source without incentives.

The Regulator Has Incentives Too

It would be convenient if government entered this story as a neutral engineer correcting a defective market. It does not.

Industries have legitimate knowledge that regulators need. A food scientist understands production constraints; a pharmacologist inside a manufacturer may know a molecule better than anyone outside it; an engineer can identify when a proposed standard is physically impossible. Excluding that expertise would produce worse rules.

The same dependence creates a route for influence. The gains from changing one regulation may be concentrated within a small industry, while the costs or benefits are dispersed in tiny amounts across millions of people. The organized side can afford specialists, sustained lobbying, legal challenges, and participation in every technical consultation. Most citizens cannot rationally spend comparable time defending their individual share of a diffuse public benefit.

Lobbying is therefore neither synonymous with corruption nor politically neutral. The OECD’s current recommendation says influence can improve policy by supplying insight and data, but warns that financially powerful actors can acquire a monopoly of influence, supply biased evidence, manipulate public opinion, and crowd out groups with fewer resources. Its remedies include disclosure, a public regulatory footprint, conflict-of-interest rules, and safeguards around the revolving door.¹⁰

This is why regulation can erode without a law being repealed. An agency may keep its formal authority while losing inspection staff, laboratory capacity, enforcement budgets, or penalties large enough to deter misconduct. An advisory process may remain open in principle while only well-funded participants can use it continuously.

The answer is not “markets or government.” Planned systems can optimize production targets while suppressing inconvenient evidence just as firms can optimize quarterly returns while externalizing harm. The recurring danger is concentrated power joined to a weak correction mechanism.

A healthier arrangement uses countervailing institutions: competition where comparison works, regulators with technical capacity, independent science, transparent lobbying, liability, antitrust, civil-society scrutiny, and democratic oversight. Each institution watches a failure mode the others are prone to miss.

Markets Need Different Rules Near the Body

Not every commercial mistake deserves the same evidentiary burden. A badly designed chair is usually inspectable, avoidable, and reversible. A subtle metabolic effect, contaminated medicine, persistent chemical, or unsafe medical device may be invisible to the buyer and appear only after exposure.

Food and healthcare combine several features that weaken ordinary consumer discipline: biological consequences, expert dependence, delayed effects, and limited ability to verify quality before purchase. Medicine adds illness, urgency, and an especially large knowledge gap between patient and provider. Food adds continuous exposure across nearly the entire population, often beginning in childhood.

That does not make food equivalent to pharmaceuticals. Regulatory burdens should follow the risk, not a slogan. The existing differences are instructive. FDA requires new drugs to establish safety and effectiveness before sale, and ordinarily requires pre-market approval for food additives. But a substance regarded by qualified experts as generally recognized as safe is excluded from the food-additive approval pathway. FDA says the scientific safety standard is meant to be equivalent, while its current notification program remains voluntary.⁵¹¹

That distinction is changing because oversight is itself a learning system. FDA’s 2026 priorities include proposing mandatory notice for new GRAS claims, building systematic post-market assessment of food chemicals, and funding research on ultra-processed foods, additives, and metabolic health.¹¹ These steps do not prove that every existing ingredient is unsafe. They show an institution recognizing that high exposure and evolving evidence require more than a one-time gate.

Competition works poorly when the costly improvement is invisible. If a safer formulation raises the price while its benefit will not become visible for twenty years, the safer producer can lose market share. A common safety floor can therefore protect competition from rewarding the firm best able to hide or export risk.

The principle is simple: the closer a market gets to human biology, and the less a buyer can detect or reverse harm, the less work “consumer choice” can be expected to do alone.

Match the Evidence Burden to the Exposure

It is impossible to demand decades of outcome data before permitting every innovation. If a technique first appears in 2026, a 30-year human study cannot exist until 2056. Waiting can also cause harm by withholding safer products, reducing food availability, or preserving a worse incumbent technology.

The sensible standard is proportional learning.

The wider the exposure, the lower the reversibility, the longer the persistence, and the more severe a plausible harm, the stronger the case for consequence research alongside deployment. Questions should include:

Reach: How many people or ecosystems will encounter the change?
Persistence: Does the product disappear, accumulate, or reproduce?
Reversibility: Can exposure be stopped and damage repaired?
Latency: Could important effects appear only after years?
System interaction: Does the intervention alter behavior, substitution, or other technologies?
Observability: Who will notice weak signals, and who must report them?
Agenda independence: Is someone funded to ask questions whose answers may reduce sales?

This framework avoids treating novelty as guilt. It treats scale as a research obligation.

The Missing Question Needs an Owner

Industrial civilization repeatedly moves through invention, optimization, adoption, visible externalities, and corrective science. The lag is not proof that science failed. Often science answered the questions its institutions rewarded.

The task is to shorten that lag without confusing precaution with paralysis. Firms should continue doing optimization and product-safety science. Regulators should specify evidence proportional to risk and remain exposed to scrutiny themselves. Public institutions should own system questions no firm can capture. Surveillance should persist after the laboratory loses sight of the product. Liability and producer responsibility should move some downstream costs into present decisions. Consumers should retain choice without being asked to perform the work of a toxicologist, epidemiologist, and regulator at the point of sale.

“Industry does science, but not where it matters” is too crude. Industry does science where its decisions live. Society’s responsibility is to make important consequences part of those decisions—and to fund the questions that will never have a natural buyer.