He’ll be presenting “Cooperation under oath: A case for context-dependent preferences” (with Joao Vaz) at 2 pm today (the link to the paper isn’t working but I’ve sent an email to try to get it fixed). Here is the “research highlights” video from his webpage:
The best part about the paper is that they are using one-sided tests, which I think is appropriate. Why do we all use two-sided tests when we have theory that signs the direction of an effect. If you tell me that it is to be more rigorous I’ll say why don’t you go with 1% (5%) significance instead of 5% (10%). As a follow-up, why do we use arbitrary cutoffs for significance levels when a p-value is continuous? We should just present the p-value (“my beta is statistically significant at the 0.087 level”). I’ve been trying to follow my own rant but sometimes old habits die hard.