Quote of the Day
Some simple things that could be done to reform Earth science from its present disaster state:
– Discard the idea of ‘p-values’ and ‘statistically significant’ data. Report likelihood functions and as much raw data as possible; have an epistemology in which different effect sizes are different hypotheses (since they make the data differently likely). A large experiment that suggests an effect size of 0.05 has *failed* to replicate a small experiment with an effect size of 0.2.
– This will also make “meta-analyses” much easier; point up when different hidden experimental conditions are producing different effect sizes and naively implying incompatible states of reality; reduce positive bias in reporting experimental results by averting the idea that an ‘insignificant’ result is a failure; etc etc. A lot of the problems in modern science really are just the result of using epistemically awful math!
Eliezer Yudkowsky @ESYudkowsky
Tweeted on April 24, 2023
This applies to more than statistics for Earth science.
Of course, I could be biased. I like working with likelihood ratios/functions.
Read the whole thing for suggestions beyond just the math.