Scott Alexander dissects a hoax chocolate study to critique common misconceptions about nutrition science, study design, and statistical methods. Longer summary

Scott Alexander analyzes a viral study claiming chocolate aids weight loss, which was revealed as a hoax designed to expose poor science journalism. He critiques four common but incorrect conclusions drawn from this incident: that people were gullible for believing it, that nutrition isn't a real science, that studies always need high sample sizes, and that p-values should be eliminated. Scott argues that there is previous research supporting chocolate's health benefits, that nutrition science uses multiple study types to build evidence, that sample size importance depends on the effect being studied, and that p-values have their place in research. He agrees with the fifth conclusion that science journalism should be trusted less, but notes that some sources like Wikipedia and specialized blogs are more reliable. Shorter summary

Scott Alexander responds to criticism of his growth mindset study analysis, acknowledging some errors while maintaining other criticisms. Longer summary

Scott Alexander responds to a critique of his previous post about a growth mindset study by Dr. Paunesku, the lead author. He acknowledges several errors in his original analysis, including misinterpreting a graph and incorrectly stating that a control group was classified as a mindset intervention. However, Scott maintains some of his criticisms, particularly regarding the combination of different interventions in the analysis and the interpretation of statistical significance. He expresses concern about loosening significance criteria and the potential for misleading conclusions when combining different interventions. Shorter summary

Scott Alexander analyzes a study revealing poor statistical literacy among doctors, critiquing both the study and its implications for medical decision-making. Longer summary

Scott Alexander discusses a study showing poor statistical literacy among doctors, particularly Ob/Gyn residents. The post highlights that only 42% of doctors correctly answered a question about p-values, and only 26% correctly solved a Bayesian probability problem about mammogram results. Scott critiques the study's questions and interpretation, notes the Dunning-Kruger effect in self-reported statistical literacy, and points out gender differences in self-assessment. He concludes by questioning the FDA's decision to restrict individuals' access to their genome information based on doctors' supposed superior statistical understanding. Shorter summary

Scott Alexander examines the claim that '90% of medical research is false', arguing it's an exaggeration while acknowledging real issues in the field. Longer summary

Scott Alexander critiques the popular claim that '90% of medical research is false', which is often attributed to John Ioannidis. He argues that this statement, while pointing to important issues, creates more panic than warranted. Scott analyzes Ioannidis' work, showing that the 90% figure is likely misinterpreted from various sources. He explains that the accuracy of medical research varies greatly depending on the type of study, with large randomized trials and meta-analyses being much more reliable. Scott also discusses how multiple studies on the same topic can greatly increase confidence in results, and how doctors' beliefs are typically based on substantial evidence rather than single studies. He concludes by acknowledging the problems in medical research while cautioning against overly cynical interpretations. Shorter summary