A few months ago I posted an essay to my website titled, “Are Political Climate Change Deniers Committing a Crime Against Humanity?” I expected to get some pushback on whether the deniers are indeed committing a crime against humanity. I was surprised, then, that the pushback I got centered instead on the scientific theory itself – that climate change is real and human-caused.
Of course, it’s possible that the commenters who “pushed back” were of the politically motivated variety – like the climate-change-denying politicians on whom the essay focused. But possibly not. Are there legitimate reasons to mistrust the science?
Certainly there are many instances in which it would be a mistake to take reported scientific results as “the truth.” I can’t remember the last time I heard of a nutritional study whose result (say, coffee is bad for you!) was not contradicted by the result of a subsequent study (coffee is good for you!).
So a healthy skepticism is a good thing. It would be a mistake to blindly believe all scientific results the way many religious people blindly believe what their religious leaders or books tell them. It would be a mistake because the results of many scientific studies turn out to be wrong. It would also be a mistake because blind belief is the exact opposite of what science demands.
But that isn’t to say we should never believe what scientists tell us. So when should we trust scientists? When should we believe that what they tell us is “the truth”?
These are actually two different, although related, questions. If we don’t trust that scientists have integrity, it will be all the more difficult to believe that what they tell us is “the truth.” So one question is whether scientists are trustworthy. But even if we think they are trustworthy, should we always believe that what they tell us is “the truth”? And if not always, when?
Scientists have to go through many years of education and typically earn advanced degrees to establish their credentials to do research. I suspect that the vast majority of scientists would have been unwilling to go through the rigors of their training if they were not truly motivated by a desire to do good science – i.e., that their motivation is to get at the truth. In contrast to many other professions, this is the hallmark of science; this is what it is about.
Yes, there are a few “bad apples.” Every once in a while we hear about some scientist who has “cooked” his data to generate some “impressive” result – and this is discovered only after his research has been published. But such cases are rare. FiveThirtyEight’s Christie Aschwanden spent several months investigating the problems in science; she does a good job of describing those problems in an article titled, “Science Isn’t Broken. It’s just a hell of a lot harder than we give it credit for.” The title encapsulates her basic takeaway.
While scientists may not be a bunch of “bad apples” out to deceive us, they are human and are therefore susceptible to forces that could impose subtle, or not so subtle, pressures on them. On occasion, scientists have been accused of “bending” to these forces in an extreme way – but these accusations tend to be politically motivated. Climate scientists have been maligned by some in the climate-change-denying crowd, who have accused the scientists of some sort of conspiracy to generate “fake” results so as to keep government research grants flowing their way. Or something. There are probably thousands of scientists working on climate change or climate change-related questions in many different countries, and there isn’t really any incentive that I can see for governments that largely fund this research to prefer one answer to another. So I personally find this conspiracy theory so implausible as to not pass the laugh test – especially coming from people who do have a vested interest in a particular answer to the question of human-caused climate change.
But are there real (subtle or not so subtle) funding influences? If an organization or industry is funding scientific research into a question whose answer will affect that funder, should we believe the results of those scientists? Before I retired, I did a lot of work for EPA on air pollution risk and benefit analyses. As part of that work I read many epidemiological studies on possible associations between various air pollutants and various adverse health effects. I couldn’t help but notice that the studies that were funded by the polluting industries tended not to find statistically significant associations between the pollutants and the health effects, while academic studies that were not funded by the polluting industries were much more likely to report statistically significant results.
This doesn’t necessarily mean there was outright fraud occurring. It could be that a subtle subconscious bias infused the industry-funded scientists’ methods. Or it could be that the funders and/or the researchers submitted only negative results – results that would not pose a problem for the industry – for possible publication in a scholarly journal.
So should we worry about the funding-related motivations of scientists? I think a healthy skepticism is warranted if the scientist or the entity funding the scientist’s research has a vested interest in a particular answer to the question being investigated.
But the greater publication bias is probably in the other direction – a bias favoring positive results – and the problem is well known in the sciences. Positive results attract more interest and readership (it’s more exciting to read that some new result was found than that nothing was found) – hence scientific journals are biased towards publishing positive findings.
At the same time, the pressure on academic researchers to ”publish or perish” encourages them to push for positive results that are more likely to get published – resulting in “p-hacking,” as described in Aschwanden’s article on FiveThirtyEight:
“These choices [that researchers make] include things like which observations to record, which ones to compare, which factors to control for … Researchers often make these calls as they go, and often there’s no obviously correct way to proceed, which makes it tempting to try different things until you get the result you’re looking for. Scientists who fiddle around like this – just about all of them do – … aren’t usually committing fraud, nor are they intending to. They’re just falling prey to natural human biases that lead them to tip the scales and set up studies to produce false-positive results. Since publishing novel results can garner a scientist rewards such as tenure and jobs, there’s ample incentive to p-hack.”
One way to circumvent the problem of publication bias might be to have studies preregistered – i.e., to have a researcher and scientific journal commit to publishing the results of a clinical trial after the study design has been determined to be adequate but before the study is actually carried out – and so before the results of the study are known. If we could compare published clinical trial results during a period without preregistration to results during a period with preregistration, such a comparison would suggest just how strong the publication bias toward positive results actually is. On the blog Neurologica, Steven Novella points to a recent study that did just that.
“The researchers looked at 30 large National Heart Lung, and Blood Institute (NHLBI) funded trials between 1970 and 2000. Of those studies, 17 or 57% showed a significant positive result. They then compared that to 25 similar studies published between 2000 and 2012. Of those, only 2 or 8% were positive. That is a significant drop – from 57% to 8% positive studies. … What was different was that starting in 2000 these trials had to be pre-registered in clinicaltrials.gov.”
This is a sobering finding – although probably not all that surprising to scientists themselves. But not all scientific studies are clinical trials. There isn’t just one “scientific method.” Scientists use different methods, depending on the science and what is feasible. The medical sciences, which do a lot of clinical trials (trying to determine if a given treatment is effective), are particularly susceptible to this problem. Evolutionary biology and climate science, which rely much more heavily on observational studies, rather than clinical trials, would be less susceptible to this kind of bias.
But perhaps not completely free of it. In 2013 over 80 scientists signed an open letter in The Guardian titled, “Trust in science would be improved by study pre-registration.” The signatory scientists suggest that something similar to the preregistration approach used for clinical trials could be applied more broadly to all kinds of scientific studies.
Especially in the social sciences, but even in the biological sciences, the systems scientists are trying to understand are extremely complex, with many variables of potential importance and many (possibly non-linear) ways they may interact with each other. It can be very difficult to collect data, let alone high quality data. Researchers often must make do with what is available and, stating the requisite caveats, proceed as best they can. And, as Aschwanden notes, researchers must make many decisions in designing their study and analyzing their data, and it’s often not obvious what the “right” decision is. These are some of the reasons we often find different studies investigating the same question coming to different conclusions.
Aschwanden describes a fascinating experiment led by Brian Nosek, a founder of the nonprofit Center for Open Science, which illustrates this point.
“Nosek’s team invited researchers to take part in a crowdsourcing data analysis project. The setup was simple. Participants were all given the same data set and prompt: Do soccer referees give more red cards to dark-skinned players than light-skinned ones? They were then asked to submit their analytical approach for feedback from other teams before diving into the analysis.
Twenty-nine teams with a total of 61 analysts took part. The researchers used a wide variety of methods … They also made different decisions about which secondary variables to use in their analyses. Despite analyzing the same data, the researchers got a variety of results. Twenty teams concluded that soccer referees gave more red cards to dark-skinned players, and nine teams found no significant relationship between skin color and red cards.
The variability in results wasn’t due to fraud or sloppy work. These were highly competent analysts who were motivated to find the truth… Even the most skilled researchers must make subjective choices that have a huge impact on the result they find.” [Emphasis added]
So one immediate takeaway is: Never put your trust in a single study! Picking (or, in some cases, cherry-picking) one study out of several can easily give a wrong conclusion. In fact, it’s not hard to see that, even if someone has found several studies that have all come to the same conclusion, we should be wary unless we know they represent a pretty complete picture of the totality of the relevant research. Or, as Scott Alexander puts it in a superb blog post over at Slate Star Codex,
“ … it’s not so much ‘beware the man of one study’ as ‘beware the man of any number of studies less than a relatively complete and not-cherry-picked survey of the research’.”
In a TED talk titled, “Why we should trust scientists,” the science historian Naomi Oreskes points out that science works by consensus. It’s not just one study or one scientist finding a result that drives where the science goes. It’s a consensus of the scientists. That’s important, because a hallmark of scientific inquiry is the rigorous process that scientific results undergo. Studies are peer reviewed and scrutinized. Scientists are not deferential to each other; they demand rigor.
Now, as Aschwanden points out, the peer review process is far from perfect. There are plenty of studies that get published that shouldn’t have been, and plenty of errors that get past the peer reviewers. (I’ve encountered some myself.) Still, the process of peer review and publication – and the demand that results be replicated in other studies – does tend to encourage rigor and scrutiny, and should ultimately result in a scientific consensus forming around those theories that are most likely to actually be true.
But isn’t it possible for a scientific consensus about a theory to develop not because of the strength of the evidence supporting it but solely as a product of a “bandwagon” phenomenon? Suppose a well-regarded scientist puts forth a theory and other scientists in that field experience pressure to agree with the theory, to “get on the bandwagon.” They may perhaps subconsciously bias their study designs in its favor. In such a case, a scientific consensus would of course be no indication at all of the validity of the theory. Does this ever happen?
Scott Alexander asked a similar question: How common are science failures? He looked for cases in which there was a scientific consensus about a theory and it turned out the theory was wrong. He limited his investigation to mistakes made in relatively recent science (1900 and later), since all sorts of mistakes were possible (and were indeed made) before scientific disciplines achieved a reasonable level of rigor. And he looked only for big mistakes – i.e., mistakes that were “part of a really important theory, one of the fundamental paradigms of an entire field” – because it is not uncommon for individual studies to report results that turn out to be “false positives.”
He came up with three examples in which a scientific consensus formed around a theory that was just plain wrong. The first example, Lysenkoism, was “the completely bonkers theory of agriculture and genetics adopted by the Soviet Union.” It had no basis in actual science, but it fit the Soviet Union’s ideology (as opposed to actual genetics, which “seemed contrary to the idea of New Soviet Man”):
“All the careerists switched over to the new paradigm, and the holdouts who continued to believe in genetics were denounced as fascists. According to Wikipedia, ‘in 1948, genetics was officially declared “a bourgeois pseudoscience”; all geneticists were fired from their jobs (some were also arrested), and all genetic research was discontinued.’”
So that’s a pretty clear example of a scientific consensus that was wrong and motivated by the scientists’ desire not to be fired from their jobs or worse – a situation that, thankfully, we do not face in the democratic Western world.
The other two examples – Freudian psychoanalysis and behaviorist psychology – are both from the field of psychology. Scott Alexander’s description of these two scientific failures is very entertaining – you should read his blog post! But what jumped out at me as I read this was that psychology is a field in which it is particularly difficult to gather actual data to test a hypothesis – especially a hypothesis like “psychoanalysis works.”
The rather striking example of Freudian psychoanalysis shows that a bandwagon effect is possible – especially if it’s all but impossible to either back up or refute the theory with hard data. In fact, I would point to that inability to support (or refute) a theory with hard data as a good reason to distrust it. But in the “hard” sciences, where it is possible to collect good data, true failures – in which there is a strong consensus that turns out to be wrong – are unlikely (at least in modern science).
Scientific inquiry is a process. Over time, studies are conducted by many different researchers. Over time, more and more data are collected and assessed. Over time, the degree of uncertainty decreases. Sometimes, particularly in the “softer” sciences, the emergence of a relatively clear picture can take a long time; and in many cases we’re still in the middle of the process – think of all the nutritional studies whose results still conflict with each other. But even given the problems that have been highlighted, the process eventually works.
The key word here is “eventually.” I must admit that I was somewhat more sanguine about scientific research when I began work on this essay than I ended up being by the time I wrote the essay. The idea that thousands of scientists around the world are conspiring to produce a false climate change picture still doesn’t pass the laugh test, and the “bad apples” in the “scientific research barrel” are relatively few, but the impact of unintentional (and intentional) biases in the scientific research and publication process, and the number of “holes” in the peer review process are apparently non-trivial.
So when should we believe what scientists tell us? There is no quick and easy answer, but here are some suggestions for how to approach the problem:
When evaluating whether I should believe the results of a particular study:
- I would start with a large dose of skepticism if the study results are what you would expect the funders of the study to want (e.g., negative results that won’t threaten an industry’s ability to continue production unhindered, or positive results that would be helpful in promoting a company’s product);
- I would read the published paper and look for obvious problems with the study design or analysis (since these can occur even in published studies). If I didn’t have sufficient expertise in the subject area to know what I was looking at, I would defer to the reviews of experts I trust.
- I would look to see how the study’s results compare with the results of other studies investigating the same (or a similar) question.
When evaluating whether I should believe a scientific theory:
- I would make sure I had a pretty complete picture of the totality of the relevant research (rather than a possibly non-representative subsample).
- I would look to see if there is corroborating evidence from other, related fields (when this is relevant).
- I would look for proposed plausible mechanisms. Studies often show significant associations between variables, but correlation does not prove causation. A plausible explanation for how one observed trend could cause another would lend support to a theory of a causal relationship.
And perhaps most important,
- I would look to see if there is a strong consensus about the scientific theory among the scientists themselves.
- And I would not accept as true any theory that is not supportable (or refutable) by hard data.
So what’s a good example of a scientific theory that we have good reason to believe? Ironically, given that it is so “hotly debated” in the political arena, the theory of human-caused climate change fits the bill quite well.
The Intergovernmental Panel on Climate Change has spent years putting together, and periodically updating, compendia of the relevant scientific literature – complete with summaries for policymakers. So for those of us who are not experts in all the relevant fields (or even one of them), this is a valuable – and, I think, trustworthy – resource that allows us to get to the “bottom line” about human-caused climate change without spending the rest of our lives poring over thousands of journal articles. And, importantly, it conveys the levels of confidence among the scientists themselves about the different aspects of climate change. About the reality of human-caused climate change, those levels of confidence are now very, very high.
The vast majority of climate change studies report results that are consistent with the theory of human-caused climate change, and there is correspondingly an overwhelming consensus among climate scientists themselves, and scientists in related fields, that climate change is indeed real and human-caused – based on mountains of data collected for over a quarter of a century.
And it’s not just climate data; there is corroborating evidence of climate change from other, related fields – and lots of it. In a world in which humans are generating greenhouse gases at a furious rate, we would expect to see not only a rising average global temperature, but the oceans becoming warmer and more acidic (as CO2 gets absorbed by the oceans and turned into carbonic acid). And that is indeed what oceanographers are finding. We would expect the world’s glaciers and ice sheets to be melting – and that is happening too. We would expect changes in the ranges of species, as temperature ranges in their original habitats change, causing them to migrate to areas to which they are better adapted; and we would expect great reductions in the populations of species that are unable to accommodate to their changing environments sufficiently quickly. We are seeing all these things. Even though there are remaining uncertainties, there is a strikingly consistent picture across many scientific disciplines – a picture of a changing climate with all the expected ramifications in their beginning stages.
Underlying all these expectations are plausible mechanisms. It would be one thing if scientists simply noticed a correlation between increasing anthropogenic greenhouse gases and increasing global average temperature. But this doesn’t prove a causal relationship. The theory of human-caused climate change is particularly solid because scientists have plausible explanations for how it is that one observed trend causes the other.
Of course, there are studies that do not support the theory of human-caused climate change and argue for alternative causes for some of the observed phenomena. Willie Soon, for example, is a prominent climate change researcher whose research purports to show that recent global warming can be explained by changes in the sun’s energy output. It turns out that Soon is funded by the fossil fuel industry (although he failed to disclose that information in many of his published papers). Other researchers funded by the fossil fuel industry tend to find either that climate change isn’t happening or that it’s not human-caused (so there’s nothing we can do about it; too bad). It would be wise to maintain a high level of skepticism about the possible funding-related biases of these scientists, since results supporting the theory of human-caused climate change would threaten the interests of their funders and thus their ability to continue their research.
If scientific inquiry is a process, then instead of framing the question as a binary choice – either we think a theory is true or we don’t – it might be more useful to think of it in a Bayesian probabilistic framework. Given the evidence to date, and given our prior beliefs about the validity of the theory, we form a (subjective) assessment of how likely it is that the theory is true. As new evidence becomes available, we reassess the likely truth of the theory.
I think this is essentially the approach scientists themselves use – as well as policymakers, who must often make policy in the face of the uncertainty born of variability in results across different studies. When evaluating the evidence for a causal relationship between a given air pollutant and a given adverse health effect, for example, EPA takes a “weight of the evidence” approach at any given point in time. At one point in time, EPA assessed the then-current literature on a possible causal relationship between ozone and premature mortality and decided that the weight of the evidence was insufficient to conclude that there is a causal relationship. At a later point in time, the weight of the evidence had shifted, in EPA’s assessment.
Thinking of the problem in this Bayesian probabilistic way also helps us to understand how even some scientific theories that have achieved a solid consensus among scientists – e.g., human-caused climate change or evolution – are still not believed by significant portions of the population. For these groups of people, there are outside factors – that have nothing to do with the validity or lack of validity of the scientific theory – that strongly influence their prior beliefs. For fundamentalist Christians, evolution is a direct challenge to their religious beliefs. For people who benefit directly or indirectly from the fossil fuel industry, the truth of human-caused climate change is a direct challenge to their near-term financial and/or political well-being. If these influences are strong enough, people’s prior beliefs about the truth of the scientific theory – in particular, their belief that it is not true – can overwhelm almost any amount of actual evidence.
Which is to say that sometimes it may not be the scientists, but we ourselves who are untrustworthy, if we have our own emotionally-, politically-, or financially-driven agenda that clouds our ability to seek truthful answers to questions for which we want particular answers.
Sometimes such an unwillingness to be truly “truth-seeking” doesn’t matter too much. A non-trivial percentage of Americans still refuse to accept the truth of evolution – a scientific theory supported by a massive amount of evidence and for which the scientific consensus is among the very strongest. But this refusal to accept the fact of evolution doesn’t do any real harm (other than cause problems in our science education and cause embarrassment for the rest of us).
Sometimes, however, an unwillingness to be truly “truth-seeking” matters a lot. The refusal by many conservatives – and particularly conservative politicians – to accept the science of human-caused climate change is doing an enormous amount of harm, because the longer we wait to address the problem, the greater the problem becomes – and we are quickly approaching the point of no return.
So the question – When should we believe scientists? – is of more than just academic interest. We can be too quick to take scientific theories as truth; or we can be too slow. And in some cases – and climate change is one of them – that can be a big deal.
 Some climate skeptics have argued that the federal government is funding climate change research as a way to get more control over our energy sources and thus more control over us. Or something. It isn’t clear how the government would have any more control over renewable energy companies than over the fossil fuel industry. Moreover, there are ways to put a price on carbon that are “revenue neutral” (lest the government increase its source of revenue with which to support the public good) – i.e., the revenue taken in by a carbon tax could ultimately be given back to the American people (e.g., by reducing other taxes). This approach has been suggested by some conservatives who are against “big government” but acknowledge the reality of human-caused climate change and the need to do something about it.
 Novella points out that “this itself is one study, although it is fairly straightforward and the results clear, but it still needs to be replicated with other databases.”
 Although Novella highlights the problem of publication bias in science, he notes that, “In the end it does work, it just takes a lot longer to get there than it should because we waste incredible resources and time chasing false positive outcomes.” See http://theness.com/neurologicablog/index.php/registering-studies-reduces-positive-outcomes/