5 ms·

>"I mean, if you spent 3-5 years (sometimes much more) of your prime working on something only to come to the conclusion that your wrong would you just accept i

by nonbel 8y ago

>"I mean, if you spent 3-5 years (sometimes much more) of your prime working on something only to come to the conclusion that your wrong would you just accept it? OR you "re-adjust" the data and somehow make your p-value significant, then BOOM, your research is valuable!" From what I've seen I don't think there is any "OR". That is standard behavior and people don't even believe there is anything wrong with it. There really is always some legitimate reason to adjust the data... the problem is that the method of drawing conclusions (checking for significant p-values) is incompatible with this. So one or the other practice needs to go.

3pt14159 8y ago
Scientists / scientists in training are starting to understand it. I think almost every last field taught in university would benefit from more statistics courses, with at least one of them from a bayesian perspective. The problem is that statistics is boring. Most things in university are causal. Statistics is kinda fuzzy and unintuitive.
- nonbel 8y ago
  I think you have the blame reversed. These problems were caused by statisticians who failed to understand there is always a reason to "adjust" the data. They taught people to apply methods that were inapplicable to the real-life research conditions they would face.
  - ethbro 8y ago
    Agreed. Statistics can be counter-intuitive, but it's not rocket science. Playing the "keeper of the true flame" and hiding the internals in intro stats courses does everyone a disservice. I get there are a huge amount of non-mathematical scientists who go through these courses, but they should at least be equipped to get a funny feeling if they see something odd (and therefore go ask for advice). Rather than just believe "that's the number that came out." Mathematically rigorous and vetting meta-studies seem like they would help here too... validating methods and that studies reproduce within a range.

by nonbel 8y ago

3pt14159 8y ago

Scientists / scientists in training are starting to understand it. I think almost every last field taught in university would benefit from more statistics courses, with at least one of them from a bayesian perspective. The problem is that statistics is boring. Most things in university are causal. Statistics is kinda fuzzy and unintuitive.

nonbel 8y ago
I think you have the blame reversed. These problems were caused by statisticians who failed to understand there is always a reason to "adjust" the data. They taught people to apply methods that were inapplicable to the real-life research conditions they would face.
- ethbro 8y ago
  Agreed. Statistics can be counter-intuitive, but it's not rocket science. Playing the "keeper of the true flame" and hiding the internals in intro stats courses does everyone a disservice. I get there are a huge amount of non-mathematical scientists who go through these courses, but they should at least be equipped to get a funny feeling if they see something odd (and therefore go ask for advice). Rather than just believe "that's the number that came out." Mathematically rigorous and vetting meta-studies seem like they would help here too... validating methods and that studies reproduce within a range.