8 ms·

Science, Now Under Scrutiny Itself

55 points by seliopou 11y ago

jugad 11y ago
The outstanding line from the article (quoting Mina Bissell)... > But it is sometimes much easier not to replicate than to replicate studies I wonder in what cases its easier to replicate studies. Probably when the study involves doing nothing...
- Al-Khwarizmi 11y ago
  In computer science, the case where the easier thing to do is to replicate something does happen sometimes. For example if you create a natural language parser that you claim is more accurate than another, you may need to replicate the other in order to compare accuracies. Sometimes you can just take the authors' reported accuracy, but often it's not that easy (maybe they just tested on English and you're interested in performance in other languages, maybe you want a different metric, maybe you want to compare on some standard set of features, etc.)
stefantalpalaru 11y ago
> It also includes moving the goal posts: that is, mining the data for results first, and then writing the paper as if the experiment had been an attempt to find just those effects. “You have exploratory findings, and you’re pitching them as ‘I knew this all along,’ as confirmatory,” Dr. Nosek said. Why is this a problem? If the experiment's design is not in conflict with the new findings, why complain?
- pessimizer 11y ago
  https://duckduckgo.com/?t=lm&q=p-hacking https://duckduckgo.com/?t=lm&q=p-hacking https://blogs.plos.org/scicomm/2015/05/19/p-hacking-megan-head-on-why-its-not-good-for-science/ https://blogs.plos.org/scicomm/2015/05/19/p-hacking-megan-he...
- imh 11y ago
  Let's say I flip a million different fair coins 20 times each. Then I analyze my findings and see that coin number 54 of them was heads all 20 times. I present my results to my peers saying, "Coin #54 is defective! The chances of this happening by chance are on in a million!" But I'd expect one of them to be heads every time flipping that many coins. It just happened to be coin #54. The problem is we're not matching up our statistical analysis to our testing when we look for findings then naively test hypotheses based on those very findings.
  - Nadya 11y ago
    "4 out of 5 [__PROFESSION__] agree. [__BRAND__] is the best!" Statistics can be true and false at the same time. Depending on interpretation of the data and how you try and 'spin it'. This is why I'm wary of any study that only shows their statistics, but does not share their testing methods.
  - ikeboy 11y ago
    Would using Bayesian statistics fix that? If the prior for a coin being defective to produce only heads is more than one in a million, then 54 is probably defective. If it's less than that, then your analysis doesn't conclude it's defective anyways. Bayes can be hacked by ommitting info, but not by things like this, I believe.
    - imh 11y ago
      It's not a bayesian versus NHST thing, it's just about doing the right test. You can do this kind of thing easily enough in a NHST kind of way. But yeah I think doing it in a bayesian manner is much easier to interpret and explain, and so requires less training, which is a HUGE boon for the crisis at hand.
      - ikeboy 11y ago
        You can do the test with NHST only if you know what the researcher had in mind when experimenting. That can lead to absurd results at times, such as your example. With Bayes, as long as no lies are told, your inference doesn't depend on the researcher's private thoughts.
- TheLoneWolfling 11y ago
  Relevant: https://xkcd.com/882/ https://xkcd.com/882/
- tjradcliffe 11y ago
  imh's explanation is good, but it's also worth considering a very common example of this. Consider two groups of people where one group engages in some activity and another does not. Measure a dozen things about those people. Then report only on the one that positively (or negatively) correlates with group membership as if it was the only thing you measured. This generates an unrealistically high statistical significance (unrealististically low p-value) because the question should not be, "What are the odds these two things are correlated by chance" but "What are the odds that any of of these things is correlated by chance with this other thing?" The significance of the result is far below the reported p-value from an analysis that assumes only one thing was investigated. You frequently see incorrect p-values of this kind even when the researchers acknowledge they tested for multiple things.
- InclinedPlane 11y ago
  Because it conflates fit and prediction. If I take a graph of the DJIA over the 20th century then fit a polynomial to it it's intellectually dishonest to turn around and claim that I have a predictive model of the stock market. I have a fitted function, nothing more. This is similar to the sharpshooter fallacy. If I shoot at the side of the barn then draw a bullseye wherever I hit I can pretend to be a good shot.
- fsloth 11y ago
  Because if one takes a large dataset with many variables and starts mining for correlations it's roughly equal to trying to get a random number generator to produce numbers in some specific pattern. More numbers, the likelier the result. This is the reason experiment repeatability should be considered of crirical importance especially for 'gooey' disciplines like medical and psychological research where there is little more than statistics to go with. Was the experiment just mininf random noise for correlation - 'simple', just repeat the experiment. If results can't be repeated it's more likely the original authors just botched the experiment, massaged their data to look how they wanted or just got lucky with a random number sequence.
petegrif 11y ago
This all seems to be 'as it should be.'Science is after all a human endeavor with all the frailties that implies. Errors and fraud are to be expected. What matters is that the process is reasonably efficient at catching such things. And the push for open data repositories and open science to increase accountability has to be a good thing.
petegrif 11y ago
This all seems to be 'as it should be.'Science is after all a human endeavor with all the frailties that implies. Errors and fraud are to be expected. What matters is that the process is reasonably efficient at catching such things. And the push for open data repositories and open science to increase accountability has to be a good thing.
hackuser 11y ago
While I think increased accountability is a good thing for the world of science, I'm concerned this will fan the anti-intellectual, anti-science rhetoric. Science, like every human institution, is flawed. That's not a reason to reject everything about it.
- ploxiln 11y ago
  It is annoying to see a retort "scientists always end up being wrong". But I don't think science should sink to the level of fake-confidence "everything going according to plan" of non-science, I think it should embrace its honesty about its mistakes. Over time people can't help but depend on science's results, and new generations take it to heart. It's actually similar to how people can't help but depend on open source tools/platforms, as much as they hated "free-tards" and said "you get what you pay for" in the past.
  - kijin 11y ago
    Well, the whole point of science is to falsify past theories and produce better theories. So of course the majority of scientific theories will end up being wrong at some point. So it's pointless to try to rebuke the "scientists always end up being wrong" like of argument. If anything, I think scientists should just bite the bullet. Theory proven wrong? That's proof that science works!
- caseysoftware 11y ago
  But it is a reason to be skeptical. Further, in recent months and years, we've learned the peer review system is completely broken. Until it's been reproduced and validated by a few outside - and hopefully disinterested - third parties, we shouldn't accept it as fact but as a theory that could be replaced.
  - andygates 11y ago
    Peer review has never been a designation of fact, just "you must be this tall to enter the fighting pits." Confusing peer review with fact is a public-perception problem, and a lot of that comes down to various antiscience groups whose work isn't even peer reviewed, let alone tested and challenged.
hackuser 11y ago
Many of those journals charge very high prices, 4 figures per year IIRC. Perhaps they could earn some of that by verifying and vouching for the integrity of what they sell.
- nycticorax 11y ago
  I think those journals charge high prices for the same reason that niche software vendors charge high prices. Namely, the product takes a goodly number of highly educated (and thus highly compensated) people to produce, and they don't have that many customers. So if the business is going to work at all, the prices have to be high. Furthermore, they are taking steps to verify the integrity of what they sell. And of course, the more steps they take, the more the journals are going to cost...
  - semi-extrinsic 11y ago
    You are aware that 98% of the highly educated people (i.e. the authors and the reviewers of papers) are not compensated at all? Sure, they're paid to do the research by their employer, but they never see a dime from the journal.
    - nycticorax 11y ago
      I am aware of that---I've written and reviewed several peer-reviewed papers. Maybe many of the journals are making money hand over fist. But I've become suspicious of arguments of the form "that seems like a high price, therefore they must have huge margins". Sometimes the price is high because the market is small.
    - vacri 11y ago
      > 98% Ah, the irony of pluck-a-number arguments when discussing this particular topic...
  - hackuser 11y ago
    I don't know about that industry, but generally prices don't depend on the sellers' costs. Prices are set to maximize profit, that is to maximize (volume * net profit). Sometimes that's a huge markup (e.g., the $5 soda at the movie theater), sometimes that works out to a loss (e.g., the 2-year-old smartphone on clearance). There are many complexities of course, such as loss-leaders, regulated markets, public goods, exploitative pricing, etc. I've read that the journals realized that they had a captive audience, i.e., that college libraries had no choice but to subscribe in order to support their faculty, and the publishers raised prices dramatically. There was a backlash from some schools at the time.
imh 11y ago
Love the article, hate the headline :( What the is wrong with these editors?
- sctb 11y ago
  We updated the title to a less bait-ey phrase from the article, and we're open to suggestions for a better one.
MarcScott 11y ago
I wonder whether a 'GitHub' for Science wouldn't help things along, just as it's helped the open source software community. Scientist could have repos with full version control, so you could go back and look at their experiments from day one. All their data would be accessible, as well as their statistical analysis, and you could see the evolution of their written paper. The repos could start off private, and then opened on the day of publishing, so others couldn't take credit for their work. Just a thought.
- foxbarrington 11y ago
  Would be even better if they opened regardless of the results. It's valuable to know what doesn't pan out / what not to try.
  - fineman 11y ago
    With an analogy to open-source, my failed projects (like we're talking about here) fall largely into two distinct categories. Those that I never started, or were totally half-baked, or turned out to trivial - the uninteresting or incomplete. 90%. And 5% turn out to be illustrative learning moments. One of my first OOP projects trying to model the real world in hierarchical classes. It was a good non-contrived example of a need for multiple orthogonal inheritance methods as any one can easily be forced into contortions in modeling even simple structures. I tried to write a project with my own trig implementations, just because. That'd at least be funny... But the rest are a few lines of drunk-coding, or something I found a library to do and never developed, or in some other way a waste of time to look at. So I don't want to hold scientists to standards I wouldn't want. Nor do I want to miss the value in the noise. So maybe we should have a honor-system somewhat like this - when you first describe to project to anyone looking for funding, or discussed it with more than a handful of peers, or ... then you make a note in your lab book. "Research direction: see if flatworms are really flat because Y - hope it'll give insight into Z". And your internal audit system submits all of these to a pool of peers who pick by interest in the area. If it's chosen, an auditor checks if the idea was ever really developed, dropping it if not, and summarizing whatever is there (or letting you write a paper) if there are results. It sounds like a lot of work, but considering that everyone is asking for grants and being audited anyways, and this would replace vast swaths of less-pointed simple bookkeeping audit with peer-review of your methods, etc. In other words, likely higher quality results for the funding bodies in less time, and with the additional benefit of being able to publish interesting failure results. Thus achieving a balance between researcher privacy and ability to try without public mockery on failure, funding efficiency, and scientific benefit.
hitchhiker999 11y ago
The rise of 'scientism' - until a few years ago I considered myself a 'science type' (excuse the phrase), immersed in logic all my life. Recently the depth of negative influence, corrupt findings, blind faith and religious-style zeal within the scientific community and pseudo scientific communities (like Reddit) has become absurd and embarrassing. Science is a method, not a conclusion. I hope it becomes 'open', and fast.
- steve19 11y ago
  Science has not changed, nor has bad science. There gave always been incentives to produce bad science, just like there has always been a moral incentive to produce good science. The religious zeal I think comes from fan boys/girls not from scientists themselves.
  - kijin 11y ago
    True, there's always been bad science. The increase in retractions can mean either that (1) there's more bad science now than there was before, or (2) we've got better at spotting bad science. Without further evidence, it's difficult to tell which one is the case. Perhaps it's a mix of both.
    - eli_gottlieb 11y ago
      I'd go with (1), caused by funding pressure. If your career can't survive when you do good science, you'll do bad science.
- mathgenius 11y ago
  Dear god of science, I pray that you will lead me to the truth, I pray that your methods, applied with utmost vigilance, will vanquish the darkness. I pray that my faith in you remains unwavering. Amen.
Tycho 11y ago
Is there any chance that there will be some sort of google-books-style centralization of experimental datasets from researchers which will then be subject to scrutiny from machine learning models trained to suss out faked data, at which point we'll uncover lots of shenanigans from our scientific past?
aaron695 11y ago
The internet is eating the world. People perhaps aren't realising that yet. Everything has changed. Just a Google search can reveal plagiarism (by accident even). A bot can find issues with made up numbers (Benford's law for starters) It's all pretty cool stuff for science.

55 points by seliopou 11y ago

jugad 11y ago

The outstanding line from the article (quoting Mina Bissell)... > But it is sometimes much easier not to replicate than to replicate studies I wonder in what cases its easier to replicate studies. Probably when the study involves doing nothing...

Al-Khwarizmi 11y ago
In computer science, the case where the easier thing to do is to replicate something does happen sometimes. For example if you create a natural language parser that you claim is more accurate than another, you may need to replicate the other in order to compare accuracies. Sometimes you can just take the authors' reported accuracy, but often it's not that easy (maybe they just tested on English and you're interested in performance in other languages, maybe you want a different metric, maybe you want to compare on some standard set of features, etc.)

stefantalpalaru 11y ago

> It also includes moving the goal posts: that is, mining the data for results first, and then writing the paper as if the experiment had been an attempt to find just those effects. “You have exploratory findings, and you’re pitching them as ‘I knew this all along,’ as confirmatory,” Dr. Nosek said. Why is this a problem? If the experiment's design is not in conflict with the new findings, why complain?

pessimizer 11y ago
https://duckduckgo.com/?t=lm&q=p-hacking https://duckduckgo.com/?t=lm&q=p-hacking https://blogs.plos.org/scicomm/2015/05/19/p-hacking-megan-head-on-why-its-not-good-for-science/ https://blogs.plos.org/scicomm/2015/05/19/p-hacking-megan-he...
imh 11y ago
Let's say I flip a million different fair coins 20 times each. Then I analyze my findings and see that coin number 54 of them was heads all 20 times. I present my results to my peers saying, "Coin #54 is defective! The chances of this happening by chance are on in a million!" But I'd expect one of them to be heads every time flipping that many coins. It just happened to be coin #54. The problem is we're not matching up our statistical analysis to our testing when we look for findings then naively test hypotheses based on those very findings.
- Nadya 11y ago
  "4 out of 5 [__PROFESSION__] agree. [__BRAND__] is the best!" Statistics can be true and false at the same time. Depending on interpretation of the data and how you try and 'spin it'. This is why I'm wary of any study that only shows their statistics, but does not share their testing methods.
- ikeboy 11y ago
  Would using Bayesian statistics fix that? If the prior for a coin being defective to produce only heads is more than one in a million, then 54 is probably defective. If it's less than that, then your analysis doesn't conclude it's defective anyways. Bayes can be hacked by ommitting info, but not by things like this, I believe.
  - imh 11y ago
    It's not a bayesian versus NHST thing, it's just about doing the right test. You can do this kind of thing easily enough in a NHST kind of way. But yeah I think doing it in a bayesian manner is much easier to interpret and explain, and so requires less training, which is a HUGE boon for the crisis at hand.
    - ikeboy 11y ago
      You can do the test with NHST only if you know what the researcher had in mind when experimenting. That can lead to absurd results at times, such as your example. With Bayes, as long as no lies are told, your inference doesn't depend on the researcher's private thoughts.
TheLoneWolfling 11y ago
Relevant: https://xkcd.com/882/ https://xkcd.com/882/
tjradcliffe 11y ago
imh's explanation is good, but it's also worth considering a very common example of this. Consider two groups of people where one group engages in some activity and another does not. Measure a dozen things about those people. Then report only on the one that positively (or negatively) correlates with group membership as if it was the only thing you measured. This generates an unrealistically high statistical significance (unrealististically low p-value) because the question should not be, "What are the odds these two things are correlated by chance" but "What are the odds that any of of these things is correlated by chance with this other thing?" The significance of the result is far below the reported p-value from an analysis that assumes only one thing was investigated. You frequently see incorrect p-values of this kind even when the researchers acknowledge they tested for multiple things.
InclinedPlane 11y ago
Because it conflates fit and prediction. If I take a graph of the DJIA over the 20th century then fit a polynomial to it it's intellectually dishonest to turn around and claim that I have a predictive model of the stock market. I have a fitted function, nothing more. This is similar to the sharpshooter fallacy. If I shoot at the side of the barn then draw a bullseye wherever I hit I can pretend to be a good shot.
fsloth 11y ago
Because if one takes a large dataset with many variables and starts mining for correlations it's roughly equal to trying to get a random number generator to produce numbers in some specific pattern. More numbers, the likelier the result. This is the reason experiment repeatability should be considered of crirical importance especially for 'gooey' disciplines like medical and psychological research where there is little more than statistics to go with. Was the experiment just mininf random noise for correlation - 'simple', just repeat the experiment. If results can't be repeated it's more likely the original authors just botched the experiment, massaged their data to look how they wanted or just got lucky with a random number sequence.

petegrif 11y ago

This all seems to be 'as it should be.'Science is after all a human endeavor with all the frailties that implies. Errors and fraud are to be expected. What matters is that the process is reasonably efficient at catching such things. And the push for open data repositories and open science to increase accountability has to be a good thing.

petegrif 11y ago

hackuser 11y ago

While I think increased accountability is a good thing for the world of science, I'm concerned this will fan the anti-intellectual, anti-science rhetoric. Science, like every human institution, is flawed. That's not a reason to reject everything about it.

ploxiln 11y ago
It is annoying to see a retort "scientists always end up being wrong". But I don't think science should sink to the level of fake-confidence "everything going according to plan" of non-science, I think it should embrace its honesty about its mistakes. Over time people can't help but depend on science's results, and new generations take it to heart. It's actually similar to how people can't help but depend on open source tools/platforms, as much as they hated "free-tards" and said "you get what you pay for" in the past.
- kijin 11y ago
  Well, the whole point of science is to falsify past theories and produce better theories. So of course the majority of scientific theories will end up being wrong at some point. So it's pointless to try to rebuke the "scientists always end up being wrong" like of argument. If anything, I think scientists should just bite the bullet. Theory proven wrong? That's proof that science works!
caseysoftware 11y ago
But it is a reason to be skeptical. Further, in recent months and years, we've learned the peer review system is completely broken. Until it's been reproduced and validated by a few outside - and hopefully disinterested - third parties, we shouldn't accept it as fact but as a theory that could be replaced.
- andygates 11y ago
  Peer review has never been a designation of fact, just "you must be this tall to enter the fighting pits." Confusing peer review with fact is a public-perception problem, and a lot of that comes down to various antiscience groups whose work isn't even peer reviewed, let alone tested and challenged.

hackuser 11y ago

Many of those journals charge very high prices, 4 figures per year IIRC. Perhaps they could earn some of that by verifying and vouching for the integrity of what they sell.

nycticorax 11y ago
I think those journals charge high prices for the same reason that niche software vendors charge high prices. Namely, the product takes a goodly number of highly educated (and thus highly compensated) people to produce, and they don't have that many customers. So if the business is going to work at all, the prices have to be high. Furthermore, they are taking steps to verify the integrity of what they sell. And of course, the more steps they take, the more the journals are going to cost...
- semi-extrinsic 11y ago
  You are aware that 98% of the highly educated people (i.e. the authors and the reviewers of papers) are not compensated at all? Sure, they're paid to do the research by their employer, but they never see a dime from the journal.
  - nycticorax 11y ago
    I am aware of that---I've written and reviewed several peer-reviewed papers. Maybe many of the journals are making money hand over fist. But I've become suspicious of arguments of the form "that seems like a high price, therefore they must have huge margins". Sometimes the price is high because the market is small.
  - vacri 11y ago
    > 98% Ah, the irony of pluck-a-number arguments when discussing this particular topic...
- hackuser 11y ago
  I don't know about that industry, but generally prices don't depend on the sellers' costs. Prices are set to maximize profit, that is to maximize (volume * net profit). Sometimes that's a huge markup (e.g., the $5 soda at the movie theater), sometimes that works out to a loss (e.g., the 2-year-old smartphone on clearance). There are many complexities of course, such as loss-leaders, regulated markets, public goods, exploitative pricing, etc. I've read that the journals realized that they had a captive audience, i.e., that college libraries had no choice but to subscribe in order to support their faculty, and the publishers raised prices dramatically. There was a backlash from some schools at the time.

imh 11y ago

Love the article, hate the headline :( What the is wrong with these editors?

sctb 11y ago
We updated the title to a less bait-ey phrase from the article, and we're open to suggestions for a better one.

MarcScott 11y ago

I wonder whether a 'GitHub' for Science wouldn't help things along, just as it's helped the open source software community. Scientist could have repos with full version control, so you could go back and look at their experiments from day one. All their data would be accessible, as well as their statistical analysis, and you could see the evolution of their written paper. The repos could start off private, and then opened on the day of publishing, so others couldn't take credit for their work. Just a thought.

foxbarrington 11y ago
Would be even better if they opened regardless of the results. It's valuable to know what doesn't pan out / what not to try.
- fineman 11y ago
  With an analogy to open-source, my failed projects (like we're talking about here) fall largely into two distinct categories. Those that I never started, or were totally half-baked, or turned out to trivial - the uninteresting or incomplete. 90%. And 5% turn out to be illustrative learning moments. One of my first OOP projects trying to model the real world in hierarchical classes. It was a good non-contrived example of a need for multiple orthogonal inheritance methods as any one can easily be forced into contortions in modeling even simple structures. I tried to write a project with my own trig implementations, just because. That'd at least be funny... But the rest are a few lines of drunk-coding, or something I found a library to do and never developed, or in some other way a waste of time to look at. So I don't want to hold scientists to standards I wouldn't want. Nor do I want to miss the value in the noise. So maybe we should have a honor-system somewhat like this - when you first describe to project to anyone looking for funding, or discussed it with more than a handful of peers, or ... then you make a note in your lab book. "Research direction: see if flatworms are really flat because Y - hope it'll give insight into Z". And your internal audit system submits all of these to a pool of peers who pick by interest in the area. If it's chosen, an auditor checks if the idea was ever really developed, dropping it if not, and summarizing whatever is there (or letting you write a paper) if there are results. It sounds like a lot of work, but considering that everyone is asking for grants and being audited anyways, and this would replace vast swaths of less-pointed simple bookkeeping audit with peer-review of your methods, etc. In other words, likely higher quality results for the funding bodies in less time, and with the additional benefit of being able to publish interesting failure results. Thus achieving a balance between researcher privacy and ability to try without public mockery on failure, funding efficiency, and scientific benefit.

hitchhiker999 11y ago

The rise of 'scientism' - until a few years ago I considered myself a 'science type' (excuse the phrase), immersed in logic all my life. Recently the depth of negative influence, corrupt findings, blind faith and religious-style zeal within the scientific community and pseudo scientific communities (like Reddit) has become absurd and embarrassing. Science is a method, not a conclusion. I hope it becomes 'open', and fast.

steve19 11y ago
Science has not changed, nor has bad science. There gave always been incentives to produce bad science, just like there has always been a moral incentive to produce good science. The religious zeal I think comes from fan boys/girls not from scientists themselves.
- kijin 11y ago
  True, there's always been bad science. The increase in retractions can mean either that (1) there's more bad science now than there was before, or (2) we've got better at spotting bad science. Without further evidence, it's difficult to tell which one is the case. Perhaps it's a mix of both.
  - eli_gottlieb 11y ago
    I'd go with (1), caused by funding pressure. If your career can't survive when you do good science, you'll do bad science.
mathgenius 11y ago
Dear god of science, I pray that you will lead me to the truth, I pray that your methods, applied with utmost vigilance, will vanquish the darkness. I pray that my faith in you remains unwavering. Amen.

Tycho 11y ago

Is there any chance that there will be some sort of google-books-style centralization of experimental datasets from researchers which will then be subject to scrutiny from machine learning models trained to suss out faked data, at which point we'll uncover lots of shenanigans from our scientific past?

aaron695 11y ago

The internet is eating the world. People perhaps aren't realising that yet. Everything has changed. Just a Google search can reveal plagiarism (by accident even). A bot can find issues with made up numbers (Benford's law for starters) It's all pretty cool stuff for science.