4 ms·

No, it's an argument against using statistics without first considering the strength of your data.

• bigfudge 11y ago
  But being a good bayesian makes you do exactly this. The process of describing priors makes it obvious you need to do a sensitivity analysis to check how much the prior is influencing the conclusions...
  • TillE 11y ago
    > being a good bayesian This is exactly what weirds people out about LessWrong folk. They talk about a tool as if it's a religion.
    • TeMPOraL 11y ago
      It's a running joke there. The people need to get over it. LW crowd is a group of people studying a pretty specific set of subjects, focused around a single website. It's typical for such a group to develop their own jargon and insider jokes, which may look weird from outside. It's normal.
    • eli_gottlieb 11y ago
      "Good Bayesian" in that context just means being an able user of Bayesian statistics, not necessarily holding any particular philosophical belief about what they mean.
• yummyfajitas 11y ago
  How can you evaluate the strength of your data without using statistics? You've created a catch-22. I'll speculate you have some sort of meta-heuristic and only apply this catch-22 under those circumstances? E.g. this catch-22 only applies to weird and socially disapproved topics?
• cobaltblue 11y ago
  Then you should take the Bayesian side, because Bayesians look at the data first, and they take their data as given rather than taking a null hypothesis as given. They don't just blindly go off and run a test (which assumes a particular prior implicitly that may be wildly inappropriate) and see what it says about the likelihood of their already observed data being generated by the test's assumed data generator.