I think mainstream science is too slow and we mere mortals can do better with Bayes.
I don’t understand what this means at all. Who are these “mere mortals” ? What are they doing, exactly, and how can they do it “better with Bayes” ? If mainstream science is too slow, what will you replace it with ?
I don’t see any available path to replacing the current system of mainstream science as performed by trained scientists with a system of bayesian science peformed by trained bayesian masters. However, comparing mainstream science to normative bayesian epistemology suggests modifications to mainstream science worth advocating for. These include reporting likelihood ratios instead of p-values, pre-registering experiments, and giving replication attempts similar prominence to the novel results they attempt to replicate. For any of these changes, the “mere mortals” doing the science will be the same trained scientist who now perform mainstream science.
It is important to remember that though there is much that Bayes says science is doing wrong, there is also much that Bayes says science is doing right.
These include reporting likelihood ratios instead of p-values, pre-registering experiments, and giving replication attempts similar prominence to the novel results they attempt to replicate.
It is possibly noteworthy that all of these are fairly mainstream positions about what sort of changes should occur. Pre-registration is already practiced in some subfields.
I suspect that pre-registering experiments and giving replication attempts similar prominence to the novel results they attempt to replicate are things that scientists who take “traditional rationality” seriously would advocate for and lament certain social structures that prevent science from being it’s ideal form. But is use of likelihood ratios instead of p-values also fairly mainstream?
Not as mainstream, but it has been discussed, particularly in the context that likelihood ratios are easier to intuitively understand what they mean. I don’t have a reference off-hand, but most versions of the argument I’ve seen advocate including both.
These are good articles, and I agree with most of what they say. However, if your thesis is that “mainstream science works too slowly, let’s throw it all out and replace it with the Bayes Rule”, then your plan is deeply flawed. There’s more to science than just p-values.
I don’t understand what this means at all. Who are these “mere mortals” ? What are they doing, exactly, and how can they do it “better with Bayes” ? If mainstream science is too slow, what will you replace it with ?
I don’t see any available path to replacing the current system of mainstream science as performed by trained scientists with a system of bayesian science peformed by trained bayesian masters. However, comparing mainstream science to normative bayesian epistemology suggests modifications to mainstream science worth advocating for. These include reporting likelihood ratios instead of p-values, pre-registering experiments, and giving replication attempts similar prominence to the novel results they attempt to replicate. For any of these changes, the “mere mortals” doing the science will be the same trained scientist who now perform mainstream science.
It is important to remember that though there is much that Bayes says science is doing wrong, there is also much that Bayes says science is doing right.
It is possibly noteworthy that all of these are fairly mainstream positions about what sort of changes should occur. Pre-registration is already practiced in some subfields.
I suspect that pre-registering experiments and giving replication attempts similar prominence to the novel results they attempt to replicate are things that scientists who take “traditional rationality” seriously would advocate for and lament certain social structures that prevent science from being it’s ideal form. But is use of likelihood ratios instead of p-values also fairly mainstream?
Not as mainstream, but it has been discussed, particularly in the context that likelihood ratios are easier to intuitively understand what they mean. I don’t have a reference off-hand, but most versions of the argument I’ve seen advocate including both.
See http://lesswrong.com/lw/1gc/frequentist_statistics_are_frequently_subjective/ and http://lesswrong.com/lw/ajj/how_to_fix_science/
These are good articles, and I agree with most of what they say. However, if your thesis is that “mainstream science works too slowly, let’s throw it all out and replace it with the Bayes Rule”, then your plan is deeply flawed. There’s more to science than just p-values.