Simulating publication bias

Research output: Working paper/Preprint Working paperResearch

214 Downloads (Pure)

Abstract

Economic research typically runs J regressions for each selected for publication – it is often selected as the ‘best’ of the regressions. The paper examines five possible meanings of the word ‘best’: SR0 is ideal selection with no bias; SR1 is polishing: selection by statistical fit; SR2 is censoring: selection by the size of estimate; SR3 selects the optimal combination of fit and size; and SR4 selects the first satisficing result. The last four SRs are steered by priors and result in bias. The MST and the FAT-PET have been developed for detection and correction of such bias. The simulations are made by data variation, while the model is the same. It appears that SR0 generates narrow funnels much at odds with observed funnels, while the other four funnels look more realistic. SR1 to SR4 give the mean a substantial bias that confirms the prior causing the bias. The FAT-PET MRA works well in finding the true value.
Original languageEnglish
Place of publicationAarhus
PublisherInstitut for Økonomi, Aarhus Universitet
Number of pages26
Publication statusPublished - 13 Dec 2013
SeriesEconomics Working Papers
Number2013-27

Cite this