Year: 2015
Type: article
Abstract: Papers in economics often try to find the ‘best’ estimate of a parameter. If researchers behave as predicted by economic theory, the research process can be modeled and simulated. The ‘best’ estimate is selected from $$J$$ regression experiments by a selection rule, SR, which gives the researcher’s preferences for the fit (significance) and size of the estimated parameter. Eight $$J$$ s and five SRs are considered. To find a stable pattern for all 40 cases, 70 million regressions are simulated. The key results are: (1) All rational SRs cause the selected estimate to be substantially biased in the direction of the priors of the researcher. (2) All such rules give almost the same bias. It can be assessed from a set of estimates, and (3) the standard PET estimate of the meta-average reduces the bias by more than 90 %.
Source: Empirical Economics
Author: Martin Paldam
Institution: Aarhus University
Language: English
Subfield: Statistics, Probability and Uncertainty
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
Funder: Deakin University
OpenAlex ID: W239369094