PSYB10H3 Lecture Notes - Lecture 10: Publication Bias, Saturated Fat, Vehicle Insurance
Document Summary
Journals do not like to publish that their research found nothing statistically significant - or nothing at all. Including statistical controls that seem defensible but were not chosen in advance: replication rates, large, multi-site projects ("many labs"), many studies chosen, original materials from authors, replication rates: 40-50%. Fixing it - publication bias: registered reports, paper published regardless of results. Journal replication sections: allows negative results to be published. Fixing it - analytic flexibility: disclosure standards, authors required to disclose dropped conditions, measures, participants, etc. If dropping any of these things, the researchers must admit to that: problems: Implications: almost certainly, some of the things we"ve been taught in class are wrong, no way to know which those are. Intergroup conflict: national popular vote, polls: clinton +3. 2%, actual: clinton +1. 7%, absolute error: 1. 5. In other words, clinton"s popular vote score was overestimated: historically, the average polling error is 2 percentage points (i. e. , the 2016 election"s polling error was below average).
