We teach pupils to be wary of information that is not backed by any study. We teach them less to be wary of the opposite argument, “the vast majority of studies show that”, which is nonetheless one of the easiest to manufacture for anyone who has the means. Here I recount what I observed, twenty years ago, of the way an industrial company handled a scientific controversy that threatened its growth, and I set this testimony alongside what researchers have established since: on the one hand, the link between the funding of studies and their results; on the other, the biases that every scientific measurement carries, even in the hands of the most rigorous and honest researchers. From this I draw what seems to me worth passing on to the audiences we want to equip in the face of information.
A festival funded by a phone operator
When we created the Pocket Films festival with the Forum des images, in 2005, the operator SFR was its main funder, and it was the first private partnership for that institution. I have told the story of the festival’s beginnings in The birth of the Pocket Films Festival (2005). This partnership brought me into close contact with the company for a few years, at a particular moment in its history: the moment when the 3G network was being rolled out, that is, when new antennas and new frequencies were being added to the old ones, and when scientists were beginning to ask about the effects of these waves on health.
I saw at close range how the company dealt with this question. The studies that concluded there was a risk were few in number, and they were a nuisance. What to do about these studies? Dispute them one by one, lie about what they said? No: the company funded others, many others, on the same subject, carried out by teams it chose, with protocols it also chose. Fairly quickly, the few studies that concluded there was a risk found themselves swamped by a mass of studies concluding that there was no risk, or only a “low” one. And it could then be said, without lying, that “the vast majority of studies” did not confirm the danger! I had no access to the contracts or the protocols, of course. I am recounting what I understood from the inside, and what I have already mentioned more briefly in Presence and « objective facts ». I saw this strategy at SFR, but there is nothing specific to that operator about it: the others did the same. Nor do I attribute to anyone in these companies an intention to lie: everyone was doing their job, which was to protect a booming business.
Fifty-nine studies sorted by funder
I would not have written this article on the strength of my memories alone. If I am writing it today, it is because a team at the University of Bern measured, two years later, the link between the funding of these studies and their results. Anke Huss, Matthias Egger and their colleagues reviewed fifty-nine experimental studies on the health effects of mobile phone use, and sorted them by source of funding (“Source of Funding and Results of Studies of Health Effects of Mobile Phone Use”, Environmental Health Perspectives, 2007). The results are clear. At least one statistically significant effect was found in a third of the studies funded exclusively by industry, and in more than four out of five of those funded by public agencies or charities. This gap does not come from the quality of the studies, since it remains when the authors take quality into account (as well as the number of outcomes each study measured). Their conclusion? To interpret a study on radiofrequencies, you need to know who paid for it.
Huss and her colleagues do not say that the waves are dangerous, but that the results of a study are statistically linked to whoever pays for it, which matches what I had seen. Four years later, in 2011, the International Agency for Research on Cancer, the cancer agency of the World Health Organization, classified radiofrequency electromagnetic fields as “possibly carcinogenic to humans”, in its Group 2B, the group of agents for which there is evidence that is not conclusive. In 2024, Ken Karipidis, of the Australian Radiation Protection and Nuclear Safety Agency, published with his colleagues a review commissioned by the WHO (Environment International, 2024). They went back over sixty-three studies published between 1994 and 2022, and found no association between mobile phone use and cancers of the head. Their work was challenged in the same journal by researchers who criticised them for underestimating the uncertainty. And in the same year, an IARC advisory group recommended re-evaluating the classification as a priority, by 2029. Fifteen years after that classification, then, we are still in uncertainty, and I do not claim to settle the matter here. For my part, I think these waves are not without effect on health. It is a conviction, I admit, and it took shape partly through seeing how people worked to make this concern inaudible.
“Bright” rats and stressed mice
Should we conclude that researchers funded by industry cheat? Of course not. Most of them do not cheat, and that is precisely what makes the phenomenon hard to see. We are going to see that what one hopes to find, the laboratory where one works, and even the smell of the person handling the animals, can change a result, without anyone wanting it or knowing it! Measuring is a science in itself, and the history of science, including the most exact sciences, is full of examples of these effects.
In the early 1960s, the American psychologist Robert Rosenthal gave laboratory rats to students whose task was to teach them to run a maze. He told half of the group that their rats had been bred for intelligence, and the other half that theirs had been bred for dullness. In reality the rats were all alike. Yet the rats reputed to be bright learned better, to a statistically significant degree, than those reputed to be dull. Nobody falsified the measurements, but when questioned afterwards, the students who believed their rats were bright said they had handled them more often and more gently than the others, without meaning to. Rosenthal published a synthesis of this work in 1963 in American Scientist, and we have spoken ever since of the experimenter expectancy effect, or the Rosenthal effect.
Physics does not escape it. In 1974, before the graduates of the California Institute of Technology, Richard Feynman gave a now famous address, “Cargo Cult Science”, in which he told the story of the measurement of the charge of the electron. Robert Millikan had measured it in the early 1910s, with a small error due to an incorrect value for the viscosity of air. Instead of correcting this error in one go, the physicists who repeated the measurement obtained values that moved away from it slowly, each a little above the previous one, until they reached the right one. Feynman gives the explanation: when a researcher obtained a result too far from Millikan’s, he looked for an error in his experiment, and found one; when he obtained a result close to it, he looked less hard. For years, on a fundamental constant, rigorous physicists thus set aside the measurements that went against their expectations, without being aware of it.
The conditions of an experiment weigh as much as expectations. In 1998, three teams of biologists, in Portland, Oregon, in Albany, New York, and in Edmonton, Canada, carried out exactly the same series of behavioural tests on the same strains of mice, of the same age, raised in the same conditions, with the same apparatus, in the same order, starting on the same day at the same local time (John Crabbe, Douglas Wahlsten and Bruce Dudek, “Genetics of Mouse Behavior: Interactions with Laboratory Environment”, Science, 1999). Yet the results differed from one laboratory to another, sometimes considerably: the mice in Edmonton were on the whole more active, one strain reacted strongly to cocaine in Portland and nowhere else, and genetically modified mice were clearly more active than the others in Portland, and no different in Edmonton. Something in each laboratory escaped even the most scrupulous standardisation. Fifteen years after the publication, Jeffrey Mogil’s team at McGill University in Montreal identified one of these things (Robert Sorge et al., Nature Methods, 2014): the presence of a male experimenter, or even of a T-shirt worn the night before by a man, is enough to stress mice as much as fifteen minutes of restraint, and makes both mice and rats less sensitive to pain. The presence of a woman does not have this effect. How many studies on pain have given different results depending on whether the person conducting them was a man or a woman? It is hard to know, since almost nobody specified it in their papers.
That is why, in serious scientific practice, the same experiments are repeated by other teams, elsewhere, blind whenever possible: a result is only considered established if it holds up under these changes. And that is why the origin of the money matters, even when everyone is honest. In a laboratory, even a university one, that works on the effects of mobile phone waves with the operators’ money, nobody needs to lie for the results to be reassuring. A study, after all, is hundreds of small decisions. How long are the cells or animals exposed, and at what power? Which population is observed? Which variations are put down to chance, and which are retained as an effect? At what point does one stop looking for an error (Feynman’s question)? If each of these decisions leans a little towards what the team expects, or towards what will help it win the next contract, the final result will lean the same way. Anyone can understand this from their own working life: consciously or not, it is very hard to reach a result that runs head-on against the interests of whoever pays you. Andreas Lundh and his colleagues have in fact measured it on a large scale, for drugs and medical devices. In a review published in 2017 by Cochrane, the international organisation that sets the standard in the evaluation of medical research (“Industry sponsorship and research outcome”), they brought together seventy-five papers. Their finding? For every hundred independent studies that conclude in favour of a product, there are roughly a hundred and thirty-four among the same number of studies funded by manufacturers. And yet, by the criteria usually checked, such as randomisation, these studies are no worse. On blinding, they were even rated well more often. The bias lies elsewhere, then, and evaluation grids cannot see it.
“Doubt is our product”
While the bias may be unconscious among researchers, the intention of manufacturers is conscious, and saying so is no accusation. When a company funds a study on its own products, it is to develop its business, which is its reason for being. One might object that the study may reveal a risk nobody suspected, and that the company will adapt. That does happen, fortunately. But imagine that an operator had to move relay antennas, lower their power, compensate people who had fallen ill… The cost would be considerable, and the company then has every interest in keeping the question “open” for as long as possible. And it has the means to do so.
The cigarette makers put it in black and white. In 1969, in an internal memo from the tobacco company Brown & Williamson, since made public along with millions of other industry documents, we read that “doubt is our product”, since it was “the best means of competing with the ’body of fact’ that exists in the mind of the general public” linking tobacco to disease. The aim was not to prove that tobacco was harmless, which was impossible, but to keep alive the idea that scientists had not made up their minds. Stéphane Foucart, a science journalist at Le Monde, traced this strategy well beyond tobacco in La Fabrique du mensonge. Comment les industriels manipulent la science et nous mettent en danger (“The Lie Factory: How Industry Manipulates Science and Puts Us in Danger”, Denoël, 2013), in the cases of asbestos, passive smoking, bisphenol A and pesticides, each time with the same lever: fund research and exploit uncertainty to delay regulation. For bisphenol A (the compound once found in particular in plastic baby bottles), Frederick vom Saal and Claude Hughes counted in 2005 the published studies on its low-dose effects in animals (Environmental Health Perspectives). There were a hundred and fifteen, and significant effects were observed in ninety-four of them. And in the eleven studies funded by industry? None!
Foucart extended this work on pesticides in Et le monde devint silencieux. Comment l’agrochimie a détruit les insectes (“And the World Fell Silent: How Agrochemicals Destroyed the Insects”, Seuil, 2019). In it he argues that neonicotinoid insecticides, introduced in the 1990s, are the main cause of the collapse of insect populations in Western Europe. What I take above all from his book is the way the manufacturers defended themselves: they gave credence to the idea that this decline was a mystery, with multiple and equivalent causes (habitat loss, disease, invasive species, climate, etc.), so that the responsibility of their products was diluted in a whole where nothing could any longer be blamed on anyone. Other products, other manufacturers, and yet I recognise in it what I had seen at SFR!
Drowning rather than denying
What can we learn from these episodes? First, that disinformation, when competent and well-funded people take charge of it, needs neither lies nor censorship. It is enough to add true studies, carried out according to the rules, often by sincere teams, until the proportion changes. Then one invokes numbers: “the majority of studies show that…”, “the scientific consensus”, “a few isolated studies”… Every word is accurate. And work that nobody has refuted becomes invisible, simply because it is drowned in the mass.
You can see the difficulty for media education. This form of disinformation looks exactly like science as it is taught at school (studies, publications, a majority, a consensus). What do we usually tell pupils? To check the source, to cross-check information, to prefer peer-reviewed publications. This advice is still sound. But imagine a pupil who cross-checks, as he has been taught, three studies funded by companies with the same interest: he gets the same conclusion three times, and comes away more convinced than before! I have shown in A Study Here, A Study There how the invocation of studies serves as an argument from authority, and in The Lies of Experts how the voice of authority takes the place of the scientific approach. The episodes I relate here show how manufacturers go about it, and research on measurement bias helps explain why it works even with scientists acting in good faith.
The questions to ask when faced with “the majority of studies”
In A Study Here, A Study There, I was already asking: who funds these studies? with what sample? what methodology? for what purpose? I suggest refining these questions in light of the episodes I have just recounted, so that you can have the participants in your workshops ask them whenever an argument rests on “the majority of studies”:
- Who funded these studies, and who had an interest in their outcome? For the waves, as we have seen, an effect appeared in a third of the studies paid for by industry, against more than four out of five of those funded by public money. For bisphenol A, none was observed in any of the eleven studies funded by industry.
- Who defined the protocols? I was already asking, in Presence and « objective facts », how the harmfulness of the waves can be measured: at what distance, with which public, over how many years? Each of these parameters is chosen by someone, and that choice steers what will be found.
- Were the results obtained blind, and reproduced by independent teams, elsewhere? Remember the mice in Portland, Albany and Edmonton, tested on the same morning, 20 April 1998, with the same apparatus: the three laboratories did not obtain the same results, and nobody had tried to deceive.
- Have the minority studies been refuted, or merely made a minority? That is not at all the same thing! In 2005, nobody had refuted the studies that troubled SFR; many others had simply been funded. And a study that nobody has refuted keeps its value, even when it stands alone.
- What do the bodies that have nothing to gain from the answer say? For the waves, the International Agency for Research on Cancer answered in 2011 “possibly carcinogenic”, and then its advisory group called for a new review in 2024. This is neither an alarm nor a clean bill of health, and it seems important to me to tell pupils exactly that.
Will these questions tell you whether the waves are dangerous? No, and that is not their purpose. They serve another purpose: to find out, when faced with “the majority of studies”, who paid and who chose the protocols. Rosenthal’s students and the physicists Feynman speaks of were not dishonest, and most scientists are no more so. They simply have, like us, budgets to balance and employers.
In 2005, thanks to SFR’s money, the Pocket Films festival lent a hundred camera phones to filmmakers such as Agnès Varda or Chris Marker, and to people who had never filmed before. At the same time, I was watching how a concern about the waves from those very phones was made inaudible. I waited twenty years to recount it in detail, for lack of proof, and I still have none. But I now know that researchers have measured the effects of this mechanism. That is why it seems important to me to talk about it with those we claim to educate about the media.
