Nov 05 2013

Is Science Broken?

Skeptics are often in a tricky position. We simultaneously are cheerleaders for science, promoting science education, scientific literacy, and the power of science as the best method for understanding the universe.

At the same time the skeptical approach requires that we explore and discuss all the various flaws, errors, and weaknesses in the institutions and process of science. Science in theory is fantastic, but it is practiced by flawed people with all their cognitive biases and perverse incentives (much like democracy or capitalism).

I think the best approach to this apparent contradiction is transparency, honesty, to be as constructive as possible, and avoid sliding into nihilism. It’s easy to focus on all the negatives about any institution, and conclude that it’s hopelessly broken. Some institutions are broken and unfixable, so it’s not an inherently unreasonable position. We should strive for a balanced and fair assessment (just like Fox news).

A recent article published in The Economist is getting a lot of play in scientific and skeptical circles. It reviews what skeptics have been talking about for years. There is a lot of crappy research out there that is unreliable. This means that just because you can find some studies that appear to support your position, it does not mean your position is correct. You cannot know the answer to a question by cherry-picking the studies you want. You have to do a critical analysis of all the research.

John Horgan blogging at Scientific American also discusses the Economist article. He adds an interesting perspective, something that is a bit of a running gag on the SGU – looking back at the hyped science news stories of the past reveals a graveyard of failed science. He discusses, for example:

One of my first assignments was profiling Jerrold S. Petrofsky, a biomedical engineer at Wright State University trying to help paralyzed patients walk by electrically stimulating their muscles with a computer-controlled device.

Horgan notes that Petrofsky’s work was overhyped and never amounted to anything, even after 30 years.

The Economist article and Horgan’s discussion are good overall, but I think a bit too negative. The full articles are worth a read, and regular readers of skeptical blogs will probably recognize many of the points and references, but will also likely learn some new details. Here is my own summary of the major areas of concern regarding the quality and reliability of published scientific research.

Most studies are of poor quality – Doing a large rigorous scientific study is difficult and requires a great deal of resources, including time and money. Researchers therefore conduct many preliminary or exploratory studies, which are small and have only basic controls. Such studies are unreliable, and are useful only in determining if further research is warranted. Preliminary studies, when positive, tend to be false positive (for reasons I will outline separately), and are much more reliable when negative.

Publish or perish – Researchers and institutions are under high pressure to publish studies. This encourages producing a high volume of low quality studies, or in publishing the “least publishable unit” of ongoing research to maximize the number of papers derived from one’s research.

Researcher bias – Researchers are people who want their ideas to be correct, and studies may be conducted by industry or those with a vested interest in the outcome. Even within accepted methodology, researchers have a great deal of degrees of freedom, or wiggle room. This freedom can be exploited (consciously or subconsciously) to engineer a positive result, even out of completely negative data. Simmons et al demonstrated that a p-value of 0.05 can be created out of negative data 60% of the time just by exploiting researcher degrees of freedom. In surveys a third of researchers admitted to engaging in questionable methods that would exploit degrees of freedom to create positive results.

Publication bias – Journals, whether in print or online, have their own motivations for success. Subscription journals want to maximize their impact factor, and that means publishing exciting studies with new and surprising findings. Such articles, of course, are the very ones that are most likely to be false positives. Open-access journals, on the other hand, that charge researchers a fee are motivated to publish lots of studies, regardless of quality. A recent Science magazine article pointed out the pathetic quality control in this segment of the industry. In addition, researchers themselves are more likely to submit a paper that is positive rather than negative.

Lack of replication – Independent replication is the key to the self-correcting nature of science. The problem is, scientists do not do replications enough, and journals do not publish them enough. There is the now famous incident of Psychology Today publishing terrible research by Daryl Bem claiming that subjects could “feel the future.” Richard Wiseman et al did an exact replication of one of Bem’s studies, with negative results, and submitted it to Psychology Today. Their response? We don’t publish exact replications. Why not? Because they are not sexy enough to boost their impact factor.

2012 review of the last century of psychology research found that only about 1% of published studies were replications. This review, as far as I can tell, has never been replicated.

Mistakes – Researchers sometimes simply make errors, and reviewers sometimes do not pick them up. A 2011 study found that 50% of neuroscience papers reviewed committed a common statistical error – an error that would often turn negative results into positive results.

Fraud – While fraud garners the most headlines, this is actually probably a small contributor to the overall problem of false positives in published research. Still, it occurs, and further contaminates the literature.

The Good

It’s not all bad, so here comes the balance.

The point of highlighting all the potential problems with research is not to argue that it’s hopeless, just that it’s difficult. We can still get to a reliable conclusion in science by carefully reviewing all the research, weeding out the bad, relying mostly on the most rigorous research that has been adequately replicated.

In other words – all of the above informs us about where to set the threshold for acceptance as scientifically proven. Skeptics tend to have a much better idea of where this threshold is than believers, who tend to use a ridiculously low threshold, at least for their preferred belief.

At its best, scientific research functions well. Researchers will carefully replicate a finding before committing their lab to furthering the research. No one wants to waste their research resources on someone else’s false positive. Early conflicting research will be vigorously debated, until a consensus protocol is agreed upon, and then all sides will listen to the results. As a result, for many important questions we have multiple replicated rigorous studies with clear results.

Also it’s important to point out that we know about all of the above problems with scientific research because scientists are asking the hard meta questions about the process of science itself. So not only is science self-correcting, the mechanisms of self-correction themselves can self-correct.


All of the problems outlined above have solutions. These solutions are not challenging or expensive, but they may be slow to be adopted because they require a culture change within science. Here are some suggestions:

Better education of researchers – most mistakes in science are just naïve and could be solved by better education. More formal and thorough education into research methodology and mistakes to avoid would help. In short, all scientists need to become better skeptics, and this is an important role for the skeptical community.

Quality control at journals – Journals need to do a better job of systematically flushing out errors and poor research quality. There are, of course, world-class science journals that do an overall excellent job of this, even though some bad studies slip through the cracks. The problem is, most journals are mediocre, and many are terrible. The journals themselves need to be better evaluated, and only those that meet a significantly high bar of quality should contribute to the official peer-reviewed published literature. We have to close the back door for bad research through bad journals.

Also, to enhance peer-review and editorial review, researchers should be required to submit all their raw data when they submit a paper for review.

Publish replications and negative studies – Journals need to set aside room for publishing negative studies and exact replications. It is selfish, in a way, for high-impact journals to take the cream of new exciting research off the top, and not do their fair share of publishing replications and negative studies. This creates a perverse incentive, where perhaps the most valuable research is neglected. For online journals, space is not an issue and therefore not a valid excuse. For print journals, sections should be created dedicated to such studies, and they can also publish online supplements with all the replications and negative studies they want, as long as they are high quality.

Register all studies – You cannot hide negative studies if you have to register them beforehand. This is already being enforced for human trials in some countries (, but other areas of research may benefit from trial registration also.

Full Disclosure – This is already largely the case, but I will include it for completeness – researchers need to fully disclose potential conflicts of interest when submitting or presenting a paper.

The media – Science journalists and educators should educate the public about the messy nature of science, and what it takes to get to a reliable conclusion. Publishing stories about preliminary research with sensational headlines hurts the public perception of science.

Advocating for improvements in the institutions of science, and public education of the methods of science, is one of the core missions for the skeptical community. The data is there; we know the problems and the solutions. We just need to apply pressure to improve what is perhaps the most important human institution – science.

25 responses so far

25 thoughts on “Is Science Broken?”

  1. oldmanjenkins says:

    The lack of negative studies published is a significant problem in pharmaceutical research. Dr Goldacre addresses this in Bad Pharma. The negative studies can even be hidden through various methods even if registered with He has taken this and has championed the AllTrials campaign. In order for an informed consent to be given by patients, the practitioners need to be aware of all results both positive and negative study results in order to make an informed decision on how to treat their patients. My father taught me there are no failures, only successes in how not to do something.

  2. dziubla says:

    While I love the idea of publishing/making available negative studies, and have advocated for it in a number of conversations, I still am not sure the best way to implement it. Is the study negative because it was poorly planned, poorly executed, or was it because it doesn’t work? Do we publish all negative studies or just the ones where we are certain the study was “properly” conducted? How do we know it was properly conducted, as sometimes the mistakes aren’t known until after the fact. Further, once negative results become available, the data mining challenge increases tremendously. I liken the problem to an paleontological dig. To some extent, publishing of negative results would be like reporting on every grain of sand that isn’t a fossil.

  3. Thanks Dr. Novella. A really fascinating case-study in the sometimes ugly, very messy, error-correcting machinery that is modern science, that I saw recently:

    Luckily people figured out what was up, and came to the conclusion that relativity wasn’t violated. But this would be a great topic for a philosophy of science class or something similar.

  4. Studies should be published based upon their quality, not their outcome. So high quality negative studies should be published, crappy studies, positive or negative, should not be.

    Preferentially publishing positive studies is data mining. It biases the literature, skewing the results of systematic reviews and meta-analysis.

    While publishing negative studies that would otherwise be neglected may increase the total number of papers published, raising the bar and only publishing quality research would decrease the total number of papers published. Doing both will reduce noise and bias, reduce data mining, and improve the literature as a whole.

  5. sonic says:

    Isn’t the notion of a ‘negative result’ a bias?

    “We found drug ‘x’ does not work for condition ‘y’ with 95% confidence, is a positive result- isn’t it? Or perhaps I should say- if the study was well done, then we can be more certain (positive) drug ‘x’ doesn’t work for condition ‘y’ than we were before.

    “This new larger study shows the earlier smaller study was in error,” is certainly a positive result- it clears up what might be an important mistake.

    I think the idea of a negative result is an example of bias.

  6. Sonic – I disagree. A positive study finds a statistically significant correlation or effect. A negative study does not – it fails to reject the null hypothesis. These outcomes are not symmetrical with regard to their statistical implications, reliability, and conclusions. The probability of false positive and false negative are not the same. Researcher bias is more likely in the positive than negative direction.

    We need to distinguish positive and negative studies. Both, however, are data and all data should be counted. That is the bias we need to eliminate.

  7. dziubla says:

    I think it can be interpreted that way, but to me negative findings are simply seeing a lack of statistical significance. I always correct students from saying “I have bad data…” Or “I have good data…”. I teach them that an experiment that informs their next question or teaches them something they didn’t know is GOOD data. even if it is something that contradicts our hypothesis or design goals.

    Steve, as for your statement regarding quality, I fully agree. However, the nature of negative findings makes it harder to publish based upon quality. There are many more easy and accurate critiques of a paper that presents a negative result than a positive result. Were the reagents confirmed? Was the reaction controlled for variable X, Y, Z, …? What cell line? What passage? What species animal? Is it appropriate? any “no” or question in those points will result in a rejection because that could explain everything. With a positive result, these questions also arise, but are easier to address with a statement of “Will look at with further studies…”

    An interesting case study on the publishing of negative results is the social sciences studies on same-sex couples and impact on child development. Much of the work has shown no statistical difference between children from same sex couples and group matched heterosexual couples. Yet, the opponents of this work have argued that the authors have selected too rigorous a p value (0.05), which will have a greater likelihood for type II error. Instead, they advocated implementing a p-value of 0.1, which will greatly increase the likelihood of type I errors, the exact errors we wish to minimize.

  8. CW says:

    At an upcoming meet-up, the science/skeptic group here in Ann Arbor of which I’m organizer, will be having a guest expert who works for the journal, Science. I’m sure some of the points listed in the summary will be brought up in our conversation.

  9. Cow_Cookie says:

    The problem is no one in the cycle has an incentive to take the steps necessary to ensure better science. Journals would suffer from lower subscriptions or paid articles. Scientists and institutions would suffer from less published work.

    The scientific community can talk all it wants about what should happen, but things won’t change unless incentives change since incentives will always trump culture. What we need is another group with competing incentives to provide checks and balances on the existing process.

    I wonder if the best thing to do would be to develop a caste of technicians whose specialty is in analyzing and replicating previous studies. They would knowingly embark on a specific career path out of grad school that would not reward unique research. Instead, it would center entirely on replication and critique of others’ research. Advancement would be based on incentives tailored to the career field — identifying methodological shortcomings, replicating studies, analyzing raw data provided by the original researchers, etc.

    This may sound like a cynical profession, but it’s one that already has counterparts in other industries. Automotive engineers don’t just work for car companies. They also work for the National Highway Traffic Safety Administration to determine when those car companies come up short.

    Of course, funding this would be a challenge. Perhaps the NIH could kick things off by earmarking a set percentage of its grants for replicative studies. Maybe a sustained lobbying effort could even convince Congress to specify that a certain portion of basic research funding must go toward such verification.

    But if there were sufficient funding for this need, institutions and associations would begin offering students the opportunity to train in the career field. Professional associations would form. Perhaps there’d even arise one or two nonprofits whose seal of approval would become the gold standard for researchers looking to prove their methodological quality.

  10. Gareth Price says:

    I agree that it is important to publish negative results when the study is attempting to replicate one where the results were positive.

    However, two things occur to me. Suppose somebody does a high quality study demonstrating that drug X has no effect on something. Nobody else is working on that topic because nobody has particular reason to believe that drug X would have any effect. Now somebody has confirmed what everyone would have suspected had they thought about it. There wouldn’t be much wrong with this but it would be a kind of easy way to get quality research into a high quality journal.

    Secondly, regarding the least publishable unit. I myself have often tried an experiment, got a negative result and pursued the matter no further. It might be useful for other people to know that. Should there be a means of publishing a very short, negative result? It seems that maybe for a negative result the least publishable unit might be smaller than for a positive one.

  11. Enzo says:

    I honestly think the solution is going to involve a centralized database where researchers can comment on articles, report replications and report contradictions. So every publication is an evolving piece, constantly being updated and interconnected to relevant work. The NIH has reportedly been working on something like this, but it can’t come soon enough.

    In the end, though, the whole system needs to be reworked. Science is way too competitive now to expect the quality of studies to improve. There needs to be a way to reward good science practices (especially for young investigators). There needs to be incentives to share data with relevant groups (and not greedily protect details that would allow replication, etc.). The journals and raw publication number need to be de-emphasized in terms of advancement. Quality score assessment needs to be stressed, etc.

  12. ccbowers says:

    “I think the best approach to this apparent contradiction is transparency, honesty, to be as constructive as possible, and avoid sliding into nihilism.”

    This extends beyond science, but any human endeavour. If someone argues that the process of science is so flawed that it is hopeless to fixed them all, then I will re-frame the situation:

    If the process is so flawed, yet we continue to make significant progress unlike any other human endeavour in our history, imagine what is possible if we make even just incremental improvements to the process. In other words, the worse you think our process is now, the more hope you should have for future improvements. For some reason, people don’t tend to think that way

  13. FacelessMan says:

    @Cow_Cookie I would suggest a different approach where students would try to replicate research as part of their BSc or MSc program (under proper supervision of course), or as an internship. I think this would also serve as a good introduction into research for the students.

    Also I agree with Enzo about the central database. From what I`ve seen many groups try replicating a study (or part of it) before starting work, that would base on the study. But when they get negative results, they just move on a differnt one, because they can`t publish. This pattern then repeats itself across many different groups around the world and a lot of time and money are wasted.

  14. Faceless Person, I am totally in agreement about having bachelor or masters theses be, at least in part, replications of previous work (especially work that is important, and interesting, to the researcher).

    Awesome idea and is what I have previously recommended to fellow students. One of the more interesting outcomes of this idea is that no replication is ever *exactly* exact anyway…

  15. Bruce says:

    In a previous life as a consultant business analyst I learned very quickly to get my fee or most of it up front for the very reason that if whatever piece of work I did came back with a “no issues” response across the board it would be assumed I had not done my job properly. Obviously, in business it is much more likely to find issues, so I can only imagine what it must be like for scientists to have to find that sexy result to justify their funding.

  16. Cow_Cookie says:

    @ FacelessMan and The Other John Mc:

    I’d submit that relegating replication to novices would only reinforce the notion that it’s not important. There may well be value in having students duplicate experiments, but the benefits are going to be more for their own learning than for establishing self-correcting mechanisms. Are students really going to be able to stand up to tenured professors on a consistent basis? There may be exceptions, such as Thomas Herndon, but there’s no way that’s going to become the rule.

  17. steve12 says:

    Enzo talked about competition and incentives above, and I cannot agree enough with this.

    Right now, the incentive structure for young investigators dictates that you should publish as much middling work as possible. In a sense, much of this is replication + boundary test. If we want to increase the pace of discovery, we need to change the funding structure to favor reasonable but higher risk work.

    This means not only having a repository of null results, but judging scientific productivity differently.

  18. steve12 says:

    “, I am totally in agreement about having bachelor or masters theses be, at least in part, replications of previous work”

    For the most part, theses at this level are at some level replications, if not straight replications.

  19. Cow Pattie, I don’t mean to say that replication should be strictly relegated to novices, but that, as steve12 said, it’s a great learning experience in that how to do the experimental design/analysis is already largely laid out, there’s a good a priori idea of what should be found, replication is accomplished, learning is accomplished, and hopefully replication+extensions are accomplished…seems like a win/win all around. But I certainly agree it is problematic that replications are unfortunately ignored or downplayed by journals, there should be more of it, there are some good ideas of how to institutionalize this, etc.

    Great discussion on all fronts though, thanks everyone!

  20. BaS says:


    In addition to your list, what about the File Drawer Effect, i.e. the fact that you can conduct 10 studies, hide the 8 negative ones and publish the 2 positive ones. This doesn’t seem to quite rise to Fraud, but it’s definitely Shenanigans and a harmful practice!

  21. daedalus2u says:

    The problem is in incentives. I think that most scientists would want to do high quality work, but they can’t for lack of resources. There are unlimited resources for activities that generate “profits”, as our capitalistic system has defined them, as in lottery fees, credit default swaps, and so on.

    If you spend your life learning how to get the most money for the least service, you will become pretty good at it. Those who spend their life learning how to do science won’t have the skills to compete for just making money.

    It isn’t that “science” is broken, it is just that our capitalistic system puts greater value on scams that make money than on good science that doesn’t.

  22. Bill Openthalt says:

    @ TOJM

    it’s a great learning experience in that how to do the experimental design/analysis is already largely laid out,

    And it brings a a fresh (call it naive if you want) view to the experiment. What the teachers have to do is to stimulate and support the questioning this lack of experience brings, to welcome the questions, and to engage the students when explaining the reasons — all too often I hear: “that’s how it’s done”.

    New brooms sweep clean.

  23. steve12 says:

    You’re exactly right daedalus2u.

    I’m pretty junior, but I feel like I’m spending most of my time chasing cash. More senior people laugh and say it only gets worse.

    And this ties into what I said above about funding safer experiments. With so few resources for basic science and no room for “failure”, the incentives are set up to fund things that will “work”.

  24. NNM says:

    There’s too much focus on string theory. It’s like a huge snowball that’s captured so many people, it won’t stop until it finds some more justifications to exist. But it’s purely a mathematical artifact with infinite possibilities.
    Physicists working on alternatives should get more respect and support.
    There are no hidden dimensions. There is no dark matter. Just particles surfing and wiggling on waves and gravitational waves covering our universe.

  25. BillyJoe7 says:

    Oh dear.

Leave a Reply