Unfortunately, the main reason these journals get used is not because they are better at selecting reliability, but because they provide authors with the credentials needed to advance their careers...
But the reason they are considered qualified to serve as credentialing authorities is, in part, due to their perceived reliability. Of course the significance of published work and its influence on the field is perhaps the primary driver of a journal's reputation for readers, but significant work can only be influential if it is also reliable. Were a journal's readership to realize that its publications could not be trusted, it would lose influence and thus the perception that having work published in it is a meaningful professional achievement.
I would like to see these sorts of analyses done over time, although I'm not sure it's possible (in the very least it would require a tremendous amount of work).
Anecdotally in speaking to people who have been in academic research for awhile, things started changing in the late 90s, plus or minus several years. I've heard a similar story repeatedly and independently about different aspects of the system, but in aggregate there's some sense that there has been change in academic research culture that started around that time.
I wonder if the relative quality signal of different types of journals has changed dramatically over time.
Is it, though? Are researchers even aware when previous research has been retracted? Retracted research often continues to amass new citations (which, of course, need not necessarily be positive).
Although I would certainly not assert that research in the "top" journals is particularly unreliable, it is not unbelievable that, when they tend to publish more extraordinary result, a smaller percentage of that tends to be true. I have not seen any indication that that leads to such journals losing influence, and it certainly doesn't lead to the Impact Factor being lowered - which often influences the "perception that having work published in it is a meaningful professional achievement".
The Retraction Watch website http://retractionwatch.com/ is one good method of highlighting retracted work. For researchers in any given field, keeping current on literature is hard work given the pace of publication but does include seeing retractions.
So, the answer to your question depends on which researchers we mean. Experts in a field have the best chance of noticing relevant retractions in their own field. If you are a researcher in one field who has interests in many other fields, it is probably much less likely that you will be able to follow the retractions in those peripheral fields. That is why Retraction Watch and other mechanisms of highlighting such errors is useful.
I've always perceived the superiority of such journals to be: a) rigorous standards for whether the paper's methodology tests their hypothesis (valid science), b) high impact of the hypothesis on the field (utility).
I don't think the OP disputes this, but it does corroborate the old saw, "To make extraordinary claims, extraordinary evidence is required".
Thus perhaps the current standards for experimentation not only fail to be extraordinary, but fundamentally are inadequate.
I don't think there's much reason to believe the "top" journals to have far more rigorous standards than the bulk of "just good" journals (which I'm sure have rigorous standards as well). It is plausible, though, that the research with potentially most impact gets submitted to the "top" journals first, resulting in the "just good" journals never having the chance to publish them.
> While the number of scientists has been growing exponentially over the last decades, the number of journals with a large audience has not kept up, neither has the number of articles published per journal. Consequently, rejection rates at the most prestigious journals has fallen below 10%
"Struggle" would imply they're actually trying. People do what is incentivized, and as long as reliability is placed towards the bottom of the priority latter, it won't improve.
It looks like the modern scientific framework just doesn't scale well. With paper output so great, it's nearly impossible for a researcher to sift through the haystack. Researchers resort to reading the papers from known groups only, which filters out most of the fluff, but perhaps leaves meaningful works from outsiders unnoticed.
Known groups. Community gossip provides sometimes-essential context. "You can't trust person X on topic Y, because when he does Z, he sees what he wants to see." Not having access to that context, is another challenge faced by folks less connected.
There are a huge number of bureaucratic hoops to jump through before you can get an article published. The flow chart is huge and gets more elaborate year by year. So if the quality is still poor then I think this simply means that we're doing it wrong. Either science isn't what we think it is or our heart isn't in it (or both).
My 'method' of choosing what to read is simple: I follow people whose work I know and enjoy. I trust them. Actually I think that's pretty standard. For example, Leonardo gets published just by having left stuff lying around the place.
I'm afraid that science indeed isn't what people usually think it is, and that there's a lot that could be improved.
Following people whose work you enjoy generally works, but it also leads to potentially valuable work by lesser-known researchers (e.g. Early-career researchers) to go unnoticed.
I think the expectations for novelty may be to high.
Look at the new drugs list every year and you see a lot of churn with minor variations common, but less novelty than many assume. https://www.centerwatch.com/drug-information/fda-approved-dr... And this is with 100+ billion in public + private money world wide per year.
Sure, there is plenty of stuff being discovered, but we can't expect a steady or increasing rate of progress across all areas.
That's the common explanation for why journals with a high Impact Factor have more retractions. Funders focus on "excellence", which is measured using the Impact Factor, which is higher when its articles are cited more, which happens more often if research is "novel" and spectacular, also known as not replicated and not in line with expectations, also known as more likely to be unreliable.
The relationship between public research funding and drug approvals is actually pretty limited. The bottleneck is actually funding for translational research, i.e. The space after public funding but before the $200B+ in big pharma r&d kicks in. This area is called the "valley of death" and is where most science dies
Venture funding is what gets drugs through the valley of death. There's a pretty good correlation between the types of drugs VC funds, the drugs big pharma companies buy from VC, and the drugs that get approved. The kinds of drugs that get approved are less correlated with areas of focus for public research spending Or diseases with most societal impact
Pharma research is not limited to the US. FDA will approve drugs developed in other countries though their will be some hoops involved they are tiny relative to finding something useful.
Anyway, as you say: "Randomly picking 24 early-stage drugs to develop is a losing bet." What I think you are missing is VC's can extract money from failed drugs. VC's are playing with other peoples money and they get a larger chunk of upside while limiting their downsides.
This means investing in VC's has poor returns on average which is why the system looks the way it does.
PS: Trying to fit that decreasing curve when the actual trend line is clearly positive is ridiculous.
Agreed that pharma research is not limited to the US, but neither is VC investing. US VC firms account for the majority of global biopharma VC investment (though china is catching up). As it stands now most FDA approved drugs are developed by US, European or Japanese firms, and in 5-10 years china will be up there as well. The US is the most profitable drug market, so as a quick and dirty analysis I think using FDA approvals, VC funding and big pharma m&a data makes sense. Not perfect, but it's a decent heuristic
VCs do make money on management fees, and VC returns in biotech from 2000-2012 were poor, but biopharma VC has done insanely well the last five years. Better returns than software VC, more IPOs and big m&a exits than tech despite only accounting for 20% of venture funding
A lot of money is now chasing these high returns, but there are not enough good biopharma entrepreneurs. So there more money but not more good startups.
And yes, that chart about r&d spending is not statistically valid :), but it isn't meant to be. But it illustrates a trend that big pharma is cutting back on r&d, especially early stage research which they outsource to startups. This topic is tangential to the topic of the post, but the article that was the source of that chart has good context around pharma r&d trends
The difficulty being you write papers to support grants and funding that some nebulous institution, or govt.dept made, to support research they want to happen, not necessarily the right research.
What do you mean by 'right research'? Proper scientific method will give reliable results no matter what motivations the scientists or their institutions have.
Do you perhaps instead have 'right' or 'wrong' priorities of science funding in mind? That's a political issue which should ultimately be decided by the voters and the representatives they choose.
Not always. There are many unknowns in research you can't control for. Some assays / models just aren't that reliable even though they are gold standard. Sometimes the equipment and reagents you use produce unexplainable results. For example, a lab I worked with was trying to make a particular type of immune cell less "aggressive". They tried all kinds of genetic modifications, hit it with different cytokines, changed its nutrients, etc, but nothing worked.
Then they used a different flask, and the results were amazing. This was a reproducible phenomenon. They had no way of knowing the equipment they used would change the immunological phenotype of their cells, but it did. There are so many confounding factors like this: reagent source, air quality in animal facilities, how you use a pipette, etc
a couple of companies are paying for research that THEY want done.
Doesn't meant that research is actually benificial for the field of study.
People from minorly related fields will be applying for this and doing research for them just to get funding, not because they have an interest in the field or in the research results.
Science isn't what most people think it is. "Reliability" is not the point of science. It never has been. This whole process started with a bunch of old white guys tooling around in dark medieval cottages trying to turn urine into gold. And yet we've still managed to come pretty far.
Science is best understood as a very special form of dialogue. It doesn't matter so much what is being said but it's very important how it is said. It is not a problem at all if one group of scientists are putting out papers claiming X and another group fails to reproduce X. That's the whole point. It's exactly how the dialogue is supposed to go. It would be a problem if one group of scientist published a paper claiming X and then somehow forbid others from trying to reproduce X. Or, when the second group failed to reproduce X the first group sought to discredit or villify or suppress the second group. This would be a big problem and would indicate a termination of the scientific process.
The real problem is that this view of science as a neverending, specialized dialogue is not understood by most people, even most scientists, unless they've been exposed to the philosophy of science. That science in itself is so poorly understood leads to silly articles about a "reproduciability crisis" (as if there was ever a time when the majority of scientific papers were shown to be reproducible) but it makes scientific claims easily impersonated ie pseudo-science.
It's not clear what the solution really is here but worrying about the "reliability" of certain papers doesn't feel like it would really move the needle.
It's OK when not reproducing something is a way of validating theories and separating valid theories from the invalid ones. It's less so when something cannot be reproduced because not enough information was shared on how it could be reproduced if it would actually have been reproducible otherwise, or if the information was shared in such a way that is it inaccessible to those who would be able to reproduce it.
The issue the article doesn't spend too much time on one of the major reasons why reliability is so low: there are too many confounding environmental variables to control for, so work done in one lab often doesn't generalize
One seemingly standard, innocuous piece of equipment can make immune cells "aggressive", while a different brand can make them tolerant. No one knows why. Genetically identical mice ordered from one vendor can consistently have different disease progression compared to mice from another vendor. Background noise in animal facilities can mess with experimental outcomes and even cure disease. Two lots of otherwise identical reagents can act almost like different chemicals
Many findings are the result of some set of confounding variables rather than the independent variable. It is impossible to know these things in advance or control for them. The best tactic is to do the same experiment in a different lab with a different research team to control for all these "other factors". Which is expensive (often six of seven figures) and can take months or even years. It is best practice when startups license tech from a university to ask for an exclusivity period where they reproduce the key experiments before committing to a full license
Imagine that for every open source library you use, all you have is a detailed read me, a set of tests, and some infrequently commented white board code. You reconstruct the code as best you can (which isn't that hard), but for some reason the tests don't check out. The bug isn't with your software. You call the original researchers and find they used a few libraries that you didn't, and these didn't show up in the publication. You have to manually install all these libraries, test them, and integrate them. Still doesn't work. You realize they used a different os and have to account for that. After six months, still nothing. After visiting their office, you realize the computer used to write the software is on the floor on some very fuzzy carpet, and there is some weird electrostatic bug in the hardware that contributed to how their software performed
This is perhaps not the best analogy, and I'm not a scientist myself though have managed scientific teams for years, managed a few tech transfers, and evaluated tons of science as potential investments, but I just wanted to drive home the point of how tough science can be. If others can critique this I would welcome it as it would be great for wet lab scientists and engineers to understand each other's worlds better
Of course there is also fraud, cherry picking of data, and careful manipulation of study design to make externalities work in your favor, but a lot of the reliability / reproducibility crisis comes down to lots of uncontrollable / unknowable variables and expensive and long development cycles
>"Many findings are the result of some set of confounding variables rather than the independent variable. It is impossible to know these things in advance or control for them."
It is literally the primary job of the (experimentalist) scientist to figure this stuff out. You are describing people skipping the science and trying to jump directly to the conclusions. This is exactly what Richard Feynman called "cargo cult science":
"For example, there have been many experiments running rats through all kinds of mazes, and so on—with little clear result. But in 1937 a man named Young did a very interesting one. He had a long corridor with doors all along one side where the rats came in, and doors along the other side where the food was. He wanted to see if he could train the rats to go in at the third door down from wherever he started them off. No. The rats went immediately to the door where the food had been the time before.
The question was, how did the rats know, because the corridor was so beautifully built and so uniform, that this was the same door as before? Obviously there was something about the door that was different from the other doors. So he painted the doors very carefully, arranging the textures on the faces of the doors exactly the same. Still the rats could tell. Then he thought maybe the rats were smelling the food, so he used chemicals to change the smell after each run. Still the rats could tell. Then he realized the rats might be able to tell by seeing the lights and the arrangement in the laboratory like any commonsense person. So he covered the corridor, and, still the rats could tell.
He finally found that they could tell by the way the floor sounded when they ran over it. And he could only fix that by putting his corridor in sand. So he covered one after another of all possible clues and finally was able to fool the rats so that they had to learn to go in the third door. If he relaxed any of his conditions, the rats could tell.
Now, from a scientific standpoint, that is an A‑Number‑l experiment. That is the experiment that makes rat‑running experiments sensible, because it uncovers the clues that the rat is really using—not what you think it’s using. And that is the experiment that tells exactly what conditions you have to use in order to be careful and control everything in an experiment with rat‑running.
I looked into the subsequent history of this research. The subsequent experiment, and the one after that, never referred to Mr. Young. They never used any of his criteria of putting the corridor on sand, or being very careful. They just went right on running rats in the same old way, and paid no attention to the great discoveries of Mr. Young, and his papers are not referred to, because he didn’t discover anything about the rats. In fact, he discovered all the things you have to do to discover something about rats. But not paying attention to experiments like that is a characteristic of Cargo Cult Science."
http://calteches.library.caltech.edu/51/2/CargoCult.htm
Do you think that many experimentalists actually are able to / have the luxury of figuring all this stuff out if an experiment works? Of course when stuff goes wrong you investigate all confounding factors until you make it work, but that doesn't mean you've chased down every loose end. And if you are doing industrial drug discovery and development, then validation, documentation and quality are absolute priorities. But from my experience (again, managing tech transfers, not being an experimentalist), people solve for getting something to work with some level of reliability, and you simply cannot test all confounding factors
To use the Feynman example: experimentalists do all of the things the rat researcher did to "debug" an experiment. But you can do the same study in a different lab, and there may be another environmental variable you have to control for that didnt exist in the first lab. Especially when you have experiments that are orders of magnitude more complex than the rat experiment
My perspective is that in theory you can do perfect science and figure out all the variables but in reality you just can't. Again, not a scientist, but I think the reproducibility issues in research and low success rates in drug development support that perspective
If you are a mathematician, there is basically no expected failure rate. You prove it or you don't. If you are an engineer, some stuff just isn't possible, and unpredictable stuff happens, but many times there's a right answer. If you are a psychologist, if you get the treatment right half the time you are good. If you are a drug researcher, even getting one approved drugs is phenomenal. It's by nature a field with low technical success rates
>"Do you think that many experimentalists actually are able to / have the luxury of figuring all this stuff out if an experiment works?"
It is a crucial part of science, not a luxury. Failure to do so means you are doing something else (I use the general term "research").
>"But you can do the same study in a different lab, and there may be another environmental variable you have to control for that didnt exist in the first lab."
Yes, independent replication is required to be sure you understand the experimental conditions.
>"My perspective is that in theory you can do perfect science and figure out all the variables but in reality you just can't... It's by nature a field with low technical success rates"
Perhaps, or perhaps this is an excuse by people who haven't even tried to actually approach the problem scientifically. They (and/or the funding agencies) just want to jump right to the exciting discoveries and cures instead of doing the science. Since the scientific approach has not been tried, we don't know.
>"the reproducibility issues in research and low success rates in drug development support that perspective"
They also support the perspective that sloppy research wastes insane amounts of time and money. Jumping to the "it is so complicated" excuse without actually trying to do things correctly seems really disingenuous.
Anyway, I left biomed exactly for this reason. There did not seem to be any interest in doing actual science. No one was ever going to try a direct replication of my work, no one cared about developing and testing quantitative models (only coming up with meaningless "significant" p-values), etc.
I agree with everything you've said. My experience comes from trying to replicate science that simply isn't reproducible in ways that are not possible to understand from reading a publication. Others I've worked with have similar experiences
These disappointing experiences with imperfect science has led me to acknowledge, as you have, that the system is not ideally set up to facilitate good science. In industrial science (pharma companies) it is better as it is harder to hide from bad science but still not perfect
I don't fault researchers for not doing perfect science. In many cases funding / resources are not sufficient to allow for exploration of all avenues. Pressure to publish is real. And as you say, independent replication seldom happens as people have incentives to spend time on other things. That is what I mean by a "luxury": if you are fighting for scarce grant dollars, or you are fighting for scarcer tenure, you can't always afford to do work like this. So there is a lot of suboptimal science. It is a sad state of affairs, but I can't blame the researchers for how the system works.
If it was as easy as writing robust software tests and having someone else run your code on their computer, there would be little excuse for bad science. But it's harder than that. I see how you can think of that as a cop out, but to me it just seems like the reflection of an imperfect reality
>"My experience comes from trying to replicate science that simply isn't reproducible in ways that are not possible to understand from reading a publication."
Right, this is because no one has done the science. Like Feynman said, someone needs to figure out all the stuff we need to keep track of to be able to do this.
I personally saw no interest in doing that (pretty basic stuff like "rat gets more food when giving this drug, therefore the drug made them smarter/healthier"; why not more hungry/motivated, etc?). It was very far from perfection. It was that someone comes up with an assay and mentions some of the limitations at the end of the paper. Then for the next 40 years no one checks this, everyone just continues to run the unverified assay and assume it measures what they want.
>"I don't fault researchers for not doing perfect science."
I just call it science vs not science, with a hard rule that in science people need to be able to independently replicate your work. Otherwise I call it research (a superset of science), and refer to the people doing it as researchers. I think this is neutral enough, while still allowing the distinction between science vs not.
Mainly, if it really is the case, I want it made clear that the current culture/environment/whatever is not allowing a scientific approach to be attempted. Maybe this other method can get somewhere, I dunno. But it does not have a proven history of success like science.
I know this may be perceived as nitpicking, but I am generally annoyed whenever the reproducibility/reliability crisis is referred to as a problem in science, writ large, rather than a specific problem in specific scientific subfields. This seems to be particularly common when the articles are written by practitioners within these subfields, such as neuroscience, biology, or psychology.
There is no reproducibility crisis in my field of science - astrophysics. To back this claim up, I searched at Retraction Watch for any mention of the top tier journals in astronomy & astrophysics - specifically, ApJ [1], MNRAS [2], A&A (none found), Icarus (none found), Nature Astronomy (none found) - and found exactly one correction and one retraction. Now certainly errata are published constantly, and I would be foolish to assume that just because Retraction Watch didn't catch many instances of scientific fraud or abuse within my field, that it practically doesn't exist. But no evidence for a "crisis" seems to exist, at least in my corner of science. Astrophysical research appears to be quite reliable and reproducible. For similar reasons, I haven't heard of a reproducibility crisis in analytical chemistry, or optical physics, or mathematics, to name a few - or, close to the interests of Hacker News, computer science. Feel free to correct me if I'm wrong.
I'm not saying this to bash non-astronomers, or non-physicists, or specifically life scientists. Biologists, neuroscientists, and psychologists are crucial participants in the greater scientific enterprise and their efforts lead more directly to alleviating human suffering than my work ever will. But when talking about reproducibility and reliability problems we cannot conflate different scientific disciplines with vastly different cultures, practices, and norms - a crisis in biology does not imply that physics has one too. Physics may have its own problems, but they aren't the same as the ones in biology.
Witness the rise of groups such as the Flat Earthers, who are rejecting basic scientific knowledge known for literal millenia.(Eratosthenes, anyone?) Or witness the anti-vaxxers, abandoning modern medicine for charlatanry. Pretending science is a monolithic enterprise and abandoning a fine-grained understanding of the validity and power of different types of scientific evidence just gives these movements strength, feeds their delusions, and weakens the prestige of scientists when we do need to stand together.
When you read between the lines of these "productivity" academic career policies and measures, you realise that they are aimed squarely at fame rather than quality of science.
I guess the argument goes something like this: when you get fame through your high profile journal publications, we will be able to attract more grants and more fee paying bums on seats.
Who cares about what you have actually written? Under the current model, nobody really.
39 comments
[ 2.8 ms ] story [ 98.3 ms ] threadEdit: See also https://medium.com/flockademic/the-ridiculous-number-that-ca...
Anecdotally in speaking to people who have been in academic research for awhile, things started changing in the late 90s, plus or minus several years. I've heard a similar story repeatedly and independently about different aspects of the system, but in aggregate there's some sense that there has been change in academic research culture that started around that time.
I wonder if the relative quality signal of different types of journals has changed dramatically over time.
Although I would certainly not assert that research in the "top" journals is particularly unreliable, it is not unbelievable that, when they tend to publish more extraordinary result, a smaller percentage of that tends to be true. I have not seen any indication that that leads to such journals losing influence, and it certainly doesn't lead to the Impact Factor being lowered - which often influences the "perception that having work published in it is a meaningful professional achievement".
So, the answer to your question depends on which researchers we mean. Experts in a field have the best chance of noticing relevant retractions in their own field. If you are a researcher in one field who has interests in many other fields, it is probably much less likely that you will be able to follow the retractions in those peripheral fields. That is why Retraction Watch and other mechanisms of highlighting such errors is useful.
I don't think the OP disputes this, but it does corroborate the old saw, "To make extraordinary claims, extraordinary evidence is required".
Thus perhaps the current standards for experimentation not only fail to be extraordinary, but fundamentally are inadequate.
> While the number of scientists has been growing exponentially over the last decades, the number of journals with a large audience has not kept up, neither has the number of articles published per journal. Consequently, rejection rates at the most prestigious journals has fallen below 10%
Do they mean acceptance rates?
Known groups. Community gossip provides sometimes-essential context. "You can't trust person X on topic Y, because when he does Z, he sees what he wants to see." Not having access to that context, is another challenge faced by folks less connected.
My 'method' of choosing what to read is simple: I follow people whose work I know and enjoy. I trust them. Actually I think that's pretty standard. For example, Leonardo gets published just by having left stuff lying around the place.
Following people whose work you enjoy generally works, but it also leads to potentially valuable work by lesser-known researchers (e.g. Early-career researchers) to go unnoticed.
Look at the new drugs list every year and you see a lot of churn with minor variations common, but less novelty than many assume. https://www.centerwatch.com/drug-information/fda-approved-dr... And this is with 100+ billion in public + private money world wide per year.
Sure, there is plenty of stuff being discovered, but we can't expect a steady or increasing rate of progress across all areas.
Venture funding is what gets drugs through the valley of death. There's a pretty good correlation between the types of drugs VC funds, the drugs big pharma companies buy from VC, and the drugs that get approved. The kinds of drugs that get approved are less correlated with areas of focus for public research spending Or diseases with most societal impact
Wrote something on this last week: http://newbio.tech/blog/vc_basics_1.html
Anyway, as you say: "Randomly picking 24 early-stage drugs to develop is a losing bet." What I think you are missing is VC's can extract money from failed drugs. VC's are playing with other peoples money and they get a larger chunk of upside while limiting their downsides.
This means investing in VC's has poor returns on average which is why the system looks the way it does.
PS: Trying to fit that decreasing curve when the actual trend line is clearly positive is ridiculous.
VCs do make money on management fees, and VC returns in biotech from 2000-2012 were poor, but biopharma VC has done insanely well the last five years. Better returns than software VC, more IPOs and big m&a exits than tech despite only accounting for 20% of venture funding
A lot of money is now chasing these high returns, but there are not enough good biopharma entrepreneurs. So there more money but not more good startups.
And yes, that chart about r&d spending is not statistically valid :), but it isn't meant to be. But it illustrates a trend that big pharma is cutting back on r&d, especially early stage research which they outsource to startups. This topic is tangential to the topic of the post, but the article that was the source of that chart has good context around pharma r&d trends
Do you perhaps instead have 'right' or 'wrong' priorities of science funding in mind? That's a political issue which should ultimately be decided by the voters and the representatives they choose.
Then they used a different flask, and the results were amazing. This was a reproducible phenomenon. They had no way of knowing the equipment they used would change the immunological phenotype of their cells, but it did. There are so many confounding factors like this: reagent source, air quality in animal facilities, how you use a pipette, etc
Doesn't mean your research is actually contributing to anything useful though.
Look at this :https://www.epsrc.ac.uk/funding/calls/irc-targeted-therapeut...
a couple of companies are paying for research that THEY want done.
Doesn't meant that research is actually benificial for the field of study.
People from minorly related fields will be applying for this and doing research for them just to get funding, not because they have an interest in the field or in the research results.
Science isn't what most people think it is. "Reliability" is not the point of science. It never has been. This whole process started with a bunch of old white guys tooling around in dark medieval cottages trying to turn urine into gold. And yet we've still managed to come pretty far.
Science is best understood as a very special form of dialogue. It doesn't matter so much what is being said but it's very important how it is said. It is not a problem at all if one group of scientists are putting out papers claiming X and another group fails to reproduce X. That's the whole point. It's exactly how the dialogue is supposed to go. It would be a problem if one group of scientist published a paper claiming X and then somehow forbid others from trying to reproduce X. Or, when the second group failed to reproduce X the first group sought to discredit or villify or suppress the second group. This would be a big problem and would indicate a termination of the scientific process.
The real problem is that this view of science as a neverending, specialized dialogue is not understood by most people, even most scientists, unless they've been exposed to the philosophy of science. That science in itself is so poorly understood leads to silly articles about a "reproduciability crisis" (as if there was ever a time when the majority of scientific papers were shown to be reproducible) but it makes scientific claims easily impersonated ie pseudo-science.
It's not clear what the solution really is here but worrying about the "reliability" of certain papers doesn't feel like it would really move the needle.
One seemingly standard, innocuous piece of equipment can make immune cells "aggressive", while a different brand can make them tolerant. No one knows why. Genetically identical mice ordered from one vendor can consistently have different disease progression compared to mice from another vendor. Background noise in animal facilities can mess with experimental outcomes and even cure disease. Two lots of otherwise identical reagents can act almost like different chemicals
Many findings are the result of some set of confounding variables rather than the independent variable. It is impossible to know these things in advance or control for them. The best tactic is to do the same experiment in a different lab with a different research team to control for all these "other factors". Which is expensive (often six of seven figures) and can take months or even years. It is best practice when startups license tech from a university to ask for an exclusivity period where they reproduce the key experiments before committing to a full license
Imagine that for every open source library you use, all you have is a detailed read me, a set of tests, and some infrequently commented white board code. You reconstruct the code as best you can (which isn't that hard), but for some reason the tests don't check out. The bug isn't with your software. You call the original researchers and find they used a few libraries that you didn't, and these didn't show up in the publication. You have to manually install all these libraries, test them, and integrate them. Still doesn't work. You realize they used a different os and have to account for that. After six months, still nothing. After visiting their office, you realize the computer used to write the software is on the floor on some very fuzzy carpet, and there is some weird electrostatic bug in the hardware that contributed to how their software performed
This is perhaps not the best analogy, and I'm not a scientist myself though have managed scientific teams for years, managed a few tech transfers, and evaluated tons of science as potential investments, but I just wanted to drive home the point of how tough science can be. If others can critique this I would welcome it as it would be great for wet lab scientists and engineers to understand each other's worlds better
Of course there is also fraud, cherry picking of data, and careful manipulation of study design to make externalities work in your favor, but a lot of the reliability / reproducibility crisis comes down to lots of uncontrollable / unknowable variables and expensive and long development cycles
It is literally the primary job of the (experimentalist) scientist to figure this stuff out. You are describing people skipping the science and trying to jump directly to the conclusions. This is exactly what Richard Feynman called "cargo cult science":
"For example, there have been many experiments running rats through all kinds of mazes, and so on—with little clear result. But in 1937 a man named Young did a very interesting one. He had a long corridor with doors all along one side where the rats came in, and doors along the other side where the food was. He wanted to see if he could train the rats to go in at the third door down from wherever he started them off. No. The rats went immediately to the door where the food had been the time before.
The question was, how did the rats know, because the corridor was so beautifully built and so uniform, that this was the same door as before? Obviously there was something about the door that was different from the other doors. So he painted the doors very carefully, arranging the textures on the faces of the doors exactly the same. Still the rats could tell. Then he thought maybe the rats were smelling the food, so he used chemicals to change the smell after each run. Still the rats could tell. Then he realized the rats might be able to tell by seeing the lights and the arrangement in the laboratory like any commonsense person. So he covered the corridor, and, still the rats could tell.
He finally found that they could tell by the way the floor sounded when they ran over it. And he could only fix that by putting his corridor in sand. So he covered one after another of all possible clues and finally was able to fool the rats so that they had to learn to go in the third door. If he relaxed any of his conditions, the rats could tell.
Now, from a scientific standpoint, that is an A‑Number‑l experiment. That is the experiment that makes rat‑running experiments sensible, because it uncovers the clues that the rat is really using—not what you think it’s using. And that is the experiment that tells exactly what conditions you have to use in order to be careful and control everything in an experiment with rat‑running.
I looked into the subsequent history of this research. The subsequent experiment, and the one after that, never referred to Mr. Young. They never used any of his criteria of putting the corridor on sand, or being very careful. They just went right on running rats in the same old way, and paid no attention to the great discoveries of Mr. Young, and his papers are not referred to, because he didn’t discover anything about the rats. In fact, he discovered all the things you have to do to discover something about rats. But not paying attention to experiments like that is a characteristic of Cargo Cult Science." http://calteches.library.caltech.edu/51/2/CargoCult.htm
To use the Feynman example: experimentalists do all of the things the rat researcher did to "debug" an experiment. But you can do the same study in a different lab, and there may be another environmental variable you have to control for that didnt exist in the first lab. Especially when you have experiments that are orders of magnitude more complex than the rat experiment
My perspective is that in theory you can do perfect science and figure out all the variables but in reality you just can't. Again, not a scientist, but I think the reproducibility issues in research and low success rates in drug development support that perspective
If you are a mathematician, there is basically no expected failure rate. You prove it or you don't. If you are an engineer, some stuff just isn't possible, and unpredictable stuff happens, but many times there's a right answer. If you are a psychologist, if you get the treatment right half the time you are good. If you are a drug researcher, even getting one approved drugs is phenomenal. It's by nature a field with low technical success rates
It is a crucial part of science, not a luxury. Failure to do so means you are doing something else (I use the general term "research").
>"But you can do the same study in a different lab, and there may be another environmental variable you have to control for that didnt exist in the first lab."
Yes, independent replication is required to be sure you understand the experimental conditions.
>"My perspective is that in theory you can do perfect science and figure out all the variables but in reality you just can't... It's by nature a field with low technical success rates"
Perhaps, or perhaps this is an excuse by people who haven't even tried to actually approach the problem scientifically. They (and/or the funding agencies) just want to jump right to the exciting discoveries and cures instead of doing the science. Since the scientific approach has not been tried, we don't know.
>"the reproducibility issues in research and low success rates in drug development support that perspective"
They also support the perspective that sloppy research wastes insane amounts of time and money. Jumping to the "it is so complicated" excuse without actually trying to do things correctly seems really disingenuous.
Anyway, I left biomed exactly for this reason. There did not seem to be any interest in doing actual science. No one was ever going to try a direct replication of my work, no one cared about developing and testing quantitative models (only coming up with meaningless "significant" p-values), etc.
These disappointing experiences with imperfect science has led me to acknowledge, as you have, that the system is not ideally set up to facilitate good science. In industrial science (pharma companies) it is better as it is harder to hide from bad science but still not perfect
I don't fault researchers for not doing perfect science. In many cases funding / resources are not sufficient to allow for exploration of all avenues. Pressure to publish is real. And as you say, independent replication seldom happens as people have incentives to spend time on other things. That is what I mean by a "luxury": if you are fighting for scarce grant dollars, or you are fighting for scarcer tenure, you can't always afford to do work like this. So there is a lot of suboptimal science. It is a sad state of affairs, but I can't blame the researchers for how the system works.
If it was as easy as writing robust software tests and having someone else run your code on their computer, there would be little excuse for bad science. But it's harder than that. I see how you can think of that as a cop out, but to me it just seems like the reflection of an imperfect reality
Right, this is because no one has done the science. Like Feynman said, someone needs to figure out all the stuff we need to keep track of to be able to do this.
I personally saw no interest in doing that (pretty basic stuff like "rat gets more food when giving this drug, therefore the drug made them smarter/healthier"; why not more hungry/motivated, etc?). It was very far from perfection. It was that someone comes up with an assay and mentions some of the limitations at the end of the paper. Then for the next 40 years no one checks this, everyone just continues to run the unverified assay and assume it measures what they want.
>"I don't fault researchers for not doing perfect science."
I just call it science vs not science, with a hard rule that in science people need to be able to independently replicate your work. Otherwise I call it research (a superset of science), and refer to the people doing it as researchers. I think this is neutral enough, while still allowing the distinction between science vs not.
Mainly, if it really is the case, I want it made clear that the current culture/environment/whatever is not allowing a scientific approach to be attempted. Maybe this other method can get somewhere, I dunno. But it does not have a proven history of success like science.
There is no reproducibility crisis in my field of science - astrophysics. To back this claim up, I searched at Retraction Watch for any mention of the top tier journals in astronomy & astrophysics - specifically, ApJ [1], MNRAS [2], A&A (none found), Icarus (none found), Nature Astronomy (none found) - and found exactly one correction and one retraction. Now certainly errata are published constantly, and I would be foolish to assume that just because Retraction Watch didn't catch many instances of scientific fraud or abuse within my field, that it practically doesn't exist. But no evidence for a "crisis" seems to exist, at least in my corner of science. Astrophysical research appears to be quite reliable and reproducible. For similar reasons, I haven't heard of a reproducibility crisis in analytical chemistry, or optical physics, or mathematics, to name a few - or, close to the interests of Hacker News, computer science. Feel free to correct me if I'm wrong.
I'm not saying this to bash non-astronomers, or non-physicists, or specifically life scientists. Biologists, neuroscientists, and psychologists are crucial participants in the greater scientific enterprise and their efforts lead more directly to alleviating human suffering than my work ever will. But when talking about reproducibility and reliability problems we cannot conflate different scientific disciplines with vastly different cultures, practices, and norms - a crisis in biology does not imply that physics has one too. Physics may have its own problems, but they aren't the same as the ones in biology.
Witness the rise of groups such as the Flat Earthers, who are rejecting basic scientific knowledge known for literal millenia.(Eratosthenes, anyone?) Or witness the anti-vaxxers, abandoning modern medicine for charlatanry. Pretending science is a monolithic enterprise and abandoning a fine-grained understanding of the validity and power of different types of scientific evidence just gives these movements strength, feeds their delusions, and weakens the prestige of scientists when we do need to stand together.
[1] http://retractionwatch.com/category/by-journal/astrophysical... [2] http://retractionwatch.com/category/by-journal/monthly-notic...
I guess the argument goes something like this: when you get fame through your high profile journal publications, we will be able to attract more grants and more fee paying bums on seats.
Who cares about what you have actually written? Under the current model, nobody really.