Scientists use a range of methods and there are no central authorities dictating which to use except what the scientific community as a whole finds interesting or helpful. That being said, Bayesian analysis is used all of the time in actual scientific papers. But by no means will you find it discussed in every paper. Doing a Bayesian analysis properly means going beyond the simple formulas you have probably seen and extending them to more complex situations (e.g. continuous possibility spaces) so it’s a significant undertaking to do this properly in a way that isn’t literally just spitballing numbers. This is just one subfield of one field that I’m somewhat familiar with but I remember seeing loads of Bayesian data analysis in cosmology. For example:
There are so many more to find. And yes it will be used a ton in medicine, materials science, etc etc.
I don’t know what advantages it would bring exaclty but I’d very much like it if we taught students to think in a Bayesian way about data, statistics, and hypothesis testing.
Bayesian statistics are sometimes very useful, but in a lot of cases they aren't of much practical use. I think it would be dumb to force a biologist to use Bayesianism when discussing the behaviour of ducks or whatever.
It's an active debate in statistical analysis, but for most scientists it doesn't make a huge practical difference.
Part of this as well is that in practice, scientists end up needing to make relatively binary decisions and conclusions, so things like “degrees of belief” just aren’t necessarily that useful much of the time (and frequentistic statistics quantities uncertainty as well, in any case). We still tend to have to decide whether we think there is a meaningful relationship or effect, whether or not to follow up on something, etc. That’s going to happen regardless of which specific type of statistical framework is used.
>If this is seen as the way forward, what governing authorities or documents in the scientific community are normatively encouraging use of Bayesianism? Or is this something scientific communities might just "do" because it's better (and mandating philosophical methods might stifle creativity?)
There is no consensus in the scientific community that Bayesian approaches are “better”, so there is no encouraging by any authorities. In fact, there is still an active discussion between Bayesian and frequentists in the field of statistics. Bayesian techniques have shown to be appropriate for some applications, but for many others the classic frequentists approaches are preferred.
>Would we see advantages if the general public (outside of scientific communities) thought more in a Bayesian way? i.e. Accepting and being more OK with uncertainty in belief as a property of carrying out inquiry?
I think so. Watch this for example:
https://youtu.be/8vHKCrNGPhY?si=2GdPdxQEG08WXr3h
Bayesian thinking is really more along the lines of, we have 7000 possibilities, each with tiny probability. some get ruled out, some get hinted against, some gain strength, some really look obvious
think about locating a crashed plane. You start with last known location and flight plans, and weather data.
at the start of a search, a map of bayesian probabilities look like the flight plan, from the last known location, smeared a bit by wind
at the end of a search, those probabilities are a map of "where they found bits"
A certain level of bayesian thought is pretty natural in the process, but not the results.
That conclusion doesn’t follow. That some statisticians think about it doesn’t mean “most scientists” do. The vast majority of non-statisticians know very little about this debate, nor do they care.
Statistics, of course, my brother recently retired from teaching that, but I assure you that NOT ONE of the practicing scientists I have ever known, in fields including biology (eg a fellow who studies bird songs), anthropology, economics, sociology, chemistry (eg food chemistry), physics, linguistics, medical research (eg a brilliant Polish scholar who studies arthritis and related issues) EVER discusses this issue at all. I am not sure any of them even know what it means even if they may have heard of it when taking some course in stat decades ago. I am sure some do but this is not what practicing scientists worry about. Maybe we should. Maybe I will mention it in something I am writing now that you have prompted me, LOL. Nor do scientists worry about Popper or Kuhn or whatever--though in some fields occasionally some such names and issues are MISused when attacking someone else, never in actually influencing one's own work. One occasonally reads an attack on some work by someone who says it is not "falsifiable".
Really? And what HAS helped science? Philosophy? Philosophy as is well known has been used in science as a blunt weapon, as I keep saying (eg Mach stopped doing science and started doing philosophy in order to do down Bolzmann and there are plenty more examples of similar abuses). But anyway the question I answered was not whether this would be desirable but whether it is in fact commonly done. Plz read the question: Bayesianism - degrees of belief - is this regularly used in science? Answer: No this is not regularly used in science.
Can science that is practiced without critically reflecting upon it's principles be considered science? Or just "common practice" as you say?
This is something Kuhn wrote about. Maybe take a gander. It certainly does explain some of the institutional blind-spots of a lot of practicing scientists that are philosophically illiterate.
Why do you insultingly suggest I don't know Kuhn's (nonsensical) claims? The question was there a particular concept is regularly discussed in science, and I answered that question. Attacking either me--or virtually all scientists--good going. Thank you and good night.
Feel like that other poster is talking about Frequentist vs. Bayesian methods (rather than philosophical interpretations of probability) which is ofcourse of practical significance to a scientist whose methods of analysis might usually fall into one or the other category. Frequential methods centered around maximum likelihood previously dominated if I am not mistaken, which I am sure is not as much of the case now. But I don't know much sbout this topic.
it may depend on the field. in a field that uses Bayesian analysis more often, like Themoopanatat mentioned for cosmology, this discussion may have been more active. I certainly remember coming across a specific paper advocsting Bayesianism in this kind of area.
I think most on this subreddit agree with me that we can't do even the most basic of science without confronting some philosophical issues.
Say we as scientists do a basic experiment with the aim of calculating acceleration of earth's gravity using a tennis ball dropped from a height. We can measure the distance, use a stopwatch etc etc.
A minimum expectation of the scientist would be to comment on the inaccuracy of the measuring devices, the inaccuracy of the observation, the simplification of our model and how it ignores aerodynamic resistance, and comment on the conclusions we can draw as to how they pertain to reality. I believe this minimum expectation of philosophical inquiry is already baked into scientific work.
We could also comment on repeating the experiment more times and how results could converge on an outcome, which I guess is a form of updating beliefs with new observation.
I think because there's no hard line between philosophy and science, the only things stopping a scientist from actually going down this rabbit hole are cost/timeframe constraints vs. objectives, and scientific community norms on the level of inquiry.
I simply say that scientists do science, not philosophy. I do not say that they are right. Of course, they should be doing philosophy, and let the sun turn around the earth until everyone understands about Bayesianism. Let people die of rabies and smallpox too.
I’m not much of a philosopher and maybe I’m not much of a scientist, either. But for what it’s worth in applied sciences like engineering and system development the Bayesian quantity is a standard part of the modeling toolkit. And those Bayesian methods produce verifiably better predictive models. Examples are applying conditional reliability when figuring failure risk and in ML models. And of course in clinical medicine there’s the diagnostic correction of the disease likelihood fallacy that famously fooled Harvard docs and med students in a 1970s study. These uses all boil down to the same Bayes probability equation using prior and posterior likelihoods.
In the early days of science, the "on or off" mode of thinking was common for a good reason. Binary truths are much more preferable than Bayesian fuzzy truth.
For instance, if you look into how scientists figured out that water is two parts hydrogen and one part oxygen, you'll see that it's completely proven to be so. It's almost impossible to believe anything else when you see the process that went into proving it. Basically they isolated oxygen and hydrogen into separate tanks, weighed them, and then introduced them together and watched the tanks empty out in a 2:1 ratio while the water weight made the difference. A lot of early science (pre-20th century) produced elegant "geometric construction"-style proofs that are observationally self-evident. Modern science (20th century and later) pretty much completely lacks this quality.
On the other hand, if you were to look into the process that went into figuring out that the Higgs-Boson exists, it's not such a simple process. It doesn't provide an elegant proof. If you really wanted to know what went into the discovery of the Higgs-Boson, it'll take you at a minimum of many years of intense studying, and even then you won't really understand it, you'll just convince yourself you get in, in order to move onto something else because reading about it is so boring and a chore to follow.
This is why I believe modern science is totally trash. The golden age of science ended in the early 20th century. Future historians will look back at this era of science and call it the "zombie science" era or something like that. Bayesian truth is a way to turn a non-result into a result so it can be published and it's author compensated.
>Are degrees of belief actually used widely and expressed in the papers of authors of modern scientific work?
I would argue that degrees of belief are actually widely used by scientists, albeit mostly implicitly and informally. (Good) Scientists will have varying degrees of certainty about specific topics, and they will update their beliefs as more data becomes available. However, it is relatively rare for this to be formally expressed in scientific papers (ymmv depending on your field, though) despite several authors promoting their use.
> I'll admit, although the formulas for calculating degree of belief were simple, actually philosophically applying them felt really challenging to apply to even simple problems. Is complexity of the Bayesian system preventing uptake?
Complexity is certainly one of the issues, but perhaps one of the biggest obstacles when it comes to adopting Bayesian methods is that of Subjectivity. It is not always trivial to quantify the degree of certainty of one's prior belief, and because priors can influence the results to the point that even weak results can be passed as substantial evidence if the priors are strong, there is concern that this could lead to 'deceptive results' from overoptimistic (or biased) researchers.
> If this is seen as the way forward, what governing authorities or documents in the scientific community are normatively encouraging use of Bayesianism? Or is this something scientific communities might just "do" because it's better (and mandating philosophical methods might stifle creativity?)
I can't speak for all fields, but as far as I know, no one is normatively encouraging the use of Bayesianism. However, some authors have suggested adopting Bayesian thinking (at least to some extent) to overcome current scientific issues such as the replicability crisis. Regina Nuzzo wrote a lovely short piece some years ago that may serve as a starting point if you want to go down this rabbit hole.
> Would we see advantages if the general public (outside of scientific communities) thought more in a Bayesian way? i.e. Accepting and being more OK with uncertainty in belief as a property of carrying out inquiry?
There is an argument that the human brain functions as a Bayesian Machine, and some authors argue that Bayesian thinking is ingrained in it. However, not everyone would agree. There is this very accessible talk by Karl Friston if you want to get the general gist. If you have time, this paper by Matteo Colombo takes a more pragmatic stance. Lastly, this commentary by Rahnev is far more critical.
In terms of the general public adopting Bayesian thinking in their daily lives, this has been widely promoted. I remember videos such as the one by Julia Galef being quite popular a few years back. There are obvious advantages to a wider adoption of Bayesian Thinking, but whether people are willing to put in the effort is another question.
Themoopanator123 | 11 hours ago
Scientists use a range of methods and there are no central authorities dictating which to use except what the scientific community as a whole finds interesting or helpful. That being said, Bayesian analysis is used all of the time in actual scientific papers. But by no means will you find it discussed in every paper. Doing a Bayesian analysis properly means going beyond the simple formulas you have probably seen and extending them to more complex situations (e.g. continuous possibility spaces) so it’s a significant undertaking to do this properly in a way that isn’t literally just spitballing numbers. This is just one subfield of one field that I’m somewhat familiar with but I remember seeing loads of Bayesian data analysis in cosmology. For example:
https://arxiv.org/abs/1906.03589,
https://arxiv.org/abs/1202.0304,
https://iopscience.iop.org/article10.1088/1475-7516/2024/04/025.
There are so many more to find. And yes it will be used a ton in medicine, materials science, etc etc.
I don’t know what advantages it would bring exaclty but I’d very much like it if we taught students to think in a Bayesian way about data, statistics, and hypothesis testing.
titotal | 9 hours ago
Bayesian statistics are sometimes very useful, but in a lot of cases they aren't of much practical use. I think it would be dumb to force a biologist to use Bayesianism when discussing the behaviour of ducks or whatever.
It's an active debate in statistical analysis, but for most scientists it doesn't make a huge practical difference.
One-Broccoli-9998 | 9 hours ago
It’s frequently used by genetic counselors to provide the probability of a child having a specific trait.
CarnivorousGoose | 6 hours ago
Part of this as well is that in practice, scientists end up needing to make relatively binary decisions and conclusions, so things like “degrees of belief” just aren’t necessarily that useful much of the time (and frequentistic statistics quantities uncertainty as well, in any case). We still tend to have to decide whether we think there is a meaningful relationship or effect, whether or not to follow up on something, etc. That’s going to happen regardless of which specific type of statistical framework is used.
Vegetable-Dust-780 | 10 hours ago
>If this is seen as the way forward, what governing authorities or documents in the scientific community are normatively encouraging use of Bayesianism? Or is this something scientific communities might just "do" because it's better (and mandating philosophical methods might stifle creativity?)
There is no consensus in the scientific community that Bayesian approaches are “better”, so there is no encouraging by any authorities. In fact, there is still an active discussion between Bayesian and frequentists in the field of statistics. Bayesian techniques have shown to be appropriate for some applications, but for many others the classic frequentists approaches are preferred.
>Would we see advantages if the general public (outside of scientific communities) thought more in a Bayesian way? i.e. Accepting and being more OK with uncertainty in belief as a property of carrying out inquiry?
I think so. Watch this for example: https://youtu.be/8vHKCrNGPhY?si=2GdPdxQEG08WXr3h
PvtRoom | 2 hours ago
Bayesian thinking is really more along the lines of, we have 7000 possibilities, each with tiny probability. some get ruled out, some get hinted against, some gain strength, some really look obvious
think about locating a crashed plane. You start with last known location and flight plans, and weather data.
at the start of a search, a map of bayesian probabilities look like the flight plan, from the last known location, smeared a bit by wind
at the end of a search, those probabilities are a map of "where they found bits"
A certain level of bayesian thought is pretty natural in the process, but not the results.
PoxonAllHoaxes | 11 hours ago
Most scientists don't think about these issues.
Vegetable-Dust-780 | 10 hours ago
The debate between Bayesian and frequentist approaches is still very actively discussed in the field of statistics, so this is is simply not true.
CarnivorousGoose | 6 hours ago
That conclusion doesn’t follow. That some statisticians think about it doesn’t mean “most scientists” do. The vast majority of non-statisticians know very little about this debate, nor do they care.
PoxonAllHoaxes | 10 hours ago
Statistics, of course, my brother recently retired from teaching that, but I assure you that NOT ONE of the practicing scientists I have ever known, in fields including biology (eg a fellow who studies bird songs), anthropology, economics, sociology, chemistry (eg food chemistry), physics, linguistics, medical research (eg a brilliant Polish scholar who studies arthritis and related issues) EVER discusses this issue at all. I am not sure any of them even know what it means even if they may have heard of it when taking some course in stat decades ago. I am sure some do but this is not what practicing scientists worry about. Maybe we should. Maybe I will mention it in something I am writing now that you have prompted me, LOL. Nor do scientists worry about Popper or Kuhn or whatever--though in some fields occasionally some such names and issues are MISused when attacking someone else, never in actually influencing one's own work. One occasonally reads an attack on some work by someone who says it is not "falsifiable".
UnderTheCurrents | 7 hours ago
Instiutionalism and "we just do it like that" has never really helped science, historically speaking.
PoxonAllHoaxes | 4 hours ago
Really? And what HAS helped science? Philosophy? Philosophy as is well known has been used in science as a blunt weapon, as I keep saying (eg Mach stopped doing science and started doing philosophy in order to do down Bolzmann and there are plenty more examples of similar abuses). But anyway the question I answered was not whether this would be desirable but whether it is in fact commonly done. Plz read the question: Bayesianism - degrees of belief - is this regularly used in science? Answer: No this is not regularly used in science.
UnderTheCurrents | 4 hours ago
I can only repeat my previous comment.
Can science that is practiced without critically reflecting upon it's principles be considered science? Or just "common practice" as you say?
This is something Kuhn wrote about. Maybe take a gander. It certainly does explain some of the institutional blind-spots of a lot of practicing scientists that are philosophically illiterate.
PoxonAllHoaxes | 3 hours ago
Why do you insultingly suggest I don't know Kuhn's (nonsensical) claims? The question was there a particular concept is regularly discussed in science, and I answered that question. Attacking either me--or virtually all scientists--good going. Thank you and good night.
UnderTheCurrents | 2 hours ago
Because you do not seem to have learned anything from them.
Can you explain to me what exactly makes his claims nonsensical?
HamiltonBrae | 6 hours ago
Feel like that other poster is talking about Frequentist vs. Bayesian methods (rather than philosophical interpretations of probability) which is ofcourse of practical significance to a scientist whose methods of analysis might usually fall into one or the other category. Frequential methods centered around maximum likelihood previously dominated if I am not mistaken, which I am sure is not as much of the case now. But I don't know much sbout this topic.
PoxonAllHoaxes | 5 hours ago
That is a good point, and I do not defend the lack of interest among scientists, but it is a fact that these things are very rarely discussed.
HamiltonBrae | 5 hours ago
it may depend on the field. in a field that uses Bayesian analysis more often, like Themoopanatat mentioned for cosmology, this discussion may have been more active. I certainly remember coming across a specific paper advocsting Bayesianism in this kind of area.
PoxonAllHoaxes | 5 hours ago
We can easily research this and see how often this occurs. I did make an allowance for occasional methodological articles in ANY science.
[OP] gnatzors | 10 hours ago
I think most on this subreddit agree with me that we can't do even the most basic of science without confronting some philosophical issues.
Say we as scientists do a basic experiment with the aim of calculating acceleration of earth's gravity using a tennis ball dropped from a height. We can measure the distance, use a stopwatch etc etc.
A minimum expectation of the scientist would be to comment on the inaccuracy of the measuring devices, the inaccuracy of the observation, the simplification of our model and how it ignores aerodynamic resistance, and comment on the conclusions we can draw as to how they pertain to reality. I believe this minimum expectation of philosophical inquiry is already baked into scientific work.
We could also comment on repeating the experiment more times and how results could converge on an outcome, which I guess is a form of updating beliefs with new observation.
I think because there's no hard line between philosophy and science, the only things stopping a scientist from actually going down this rabbit hole are cost/timeframe constraints vs. objectives, and scientific community norms on the level of inquiry.
PoxonAllHoaxes | 10 hours ago
I simply say that scientists do science, not philosophy. I do not say that they are right. Of course, they should be doing philosophy, and let the sun turn around the earth until everyone understands about Bayesianism. Let people die of rabies and smallpox too.
BoneSpring | 7 hours ago
And neither scientists nor statisticians use "belief" to frame their conclusions.
radiodigm | 5 hours ago
I’m not much of a philosopher and maybe I’m not much of a scientist, either. But for what it’s worth in applied sciences like engineering and system development the Bayesian quantity is a standard part of the modeling toolkit. And those Bayesian methods produce verifiably better predictive models. Examples are applying conditional reliability when figuring failure risk and in ML models. And of course in clinical medicine there’s the diagnostic correction of the disease likelihood fallacy that famously fooled Harvard docs and med students in a 1970s study. These uses all boil down to the same Bayes probability equation using prior and posterior likelihoods.
freework | 4 hours ago
In the early days of science, the "on or off" mode of thinking was common for a good reason. Binary truths are much more preferable than Bayesian fuzzy truth.
For instance, if you look into how scientists figured out that water is two parts hydrogen and one part oxygen, you'll see that it's completely proven to be so. It's almost impossible to believe anything else when you see the process that went into proving it. Basically they isolated oxygen and hydrogen into separate tanks, weighed them, and then introduced them together and watched the tanks empty out in a 2:1 ratio while the water weight made the difference. A lot of early science (pre-20th century) produced elegant "geometric construction"-style proofs that are observationally self-evident. Modern science (20th century and later) pretty much completely lacks this quality.
On the other hand, if you were to look into the process that went into figuring out that the Higgs-Boson exists, it's not such a simple process. It doesn't provide an elegant proof. If you really wanted to know what went into the discovery of the Higgs-Boson, it'll take you at a minimum of many years of intense studying, and even then you won't really understand it, you'll just convince yourself you get in, in order to move onto something else because reading about it is so boring and a chore to follow.
This is why I believe modern science is totally trash. The golden age of science ended in the early 20th century. Future historians will look back at this era of science and call it the "zombie science" era or something like that. Bayesian truth is a way to turn a non-result into a result so it can be published and it's author compensated.
DrPapaDragonX13 | 30 minutes ago
>Are degrees of belief actually used widely and expressed in the papers of authors of modern scientific work?
I would argue that degrees of belief are actually widely used by scientists, albeit mostly implicitly and informally. (Good) Scientists will have varying degrees of certainty about specific topics, and they will update their beliefs as more data becomes available. However, it is relatively rare for this to be formally expressed in scientific papers (ymmv depending on your field, though) despite several authors promoting their use.
> I'll admit, although the formulas for calculating degree of belief were simple, actually philosophically applying them felt really challenging to apply to even simple problems. Is complexity of the Bayesian system preventing uptake?
Complexity is certainly one of the issues, but perhaps one of the biggest obstacles when it comes to adopting Bayesian methods is that of Subjectivity. It is not always trivial to quantify the degree of certainty of one's prior belief, and because priors can influence the results to the point that even weak results can be passed as substantial evidence if the priors are strong, there is concern that this could lead to 'deceptive results' from overoptimistic (or biased) researchers.
> If this is seen as the way forward, what governing authorities or documents in the scientific community are normatively encouraging use of Bayesianism? Or is this something scientific communities might just "do" because it's better (and mandating philosophical methods might stifle creativity?)
I can't speak for all fields, but as far as I know, no one is normatively encouraging the use of Bayesianism. However, some authors have suggested adopting Bayesian thinking (at least to some extent) to overcome current scientific issues such as the replicability crisis. Regina Nuzzo wrote a lovely short piece some years ago that may serve as a starting point if you want to go down this rabbit hole.
> Would we see advantages if the general public (outside of scientific communities) thought more in a Bayesian way? i.e. Accepting and being more OK with uncertainty in belief as a property of carrying out inquiry?
There is an argument that the human brain functions as a Bayesian Machine, and some authors argue that Bayesian thinking is ingrained in it. However, not everyone would agree. There is this very accessible talk by Karl Friston if you want to get the general gist. If you have time, this paper by Matteo Colombo takes a more pragmatic stance. Lastly, this commentary by Rahnev is far more critical.
In terms of the general public adopting Bayesian thinking in their daily lives, this has been widely promoted. I remember videos such as the one by Julia Galef being quite popular a few years back. There are obvious advantages to a wider adoption of Bayesian Thinking, but whether people are willing to put in the effort is another question.