Within multiple fields alarming reproducibility problems are now obvious to most: The majority of the reported effects are either false positives or the population effect size is much smaller than expected based on the initial studies (e.g., Ioannidis, 2005; Button et al., 2013; Open Science Collaboration, 2015; Baker, 2016; Nichols et al., 2017). Assuming that neither outright scientific fraud (Fanelli, 2009) nor severe deficits in methodological training are the norm, likely reasons for this inacceptable status quo include the following: (A) a high prevalence of severely underpowered studies (e.g., Button et al., 2013), (B) hypothesizing after results are known (HARKing; Kerr, 1998), (C) intentionally or unintentionally exploiting researcher degrees of freedom (Simmons et al., 2011) in data processing and analysis and thereby pushing the p-value of statistical tests below the conventional significance level without being transparent concerning all the variables and approaches that have been tried out (P-HACKING), and (D) selective reporting of research findings and publication bias. Several options for pre-registration of hypotheses are now readily available providing the opportunity to effectively prevent HARKing (e.g., OSF.io, AsPredicted.org). However, suggestions to address the other three issues have so far met with the following challenges:.
展开▼
