Wolf
Vanpaemel, professor of Quantitative
Psychology, talks about transparency for
better and reliable research. He shows what researchers can do to ensure readers
know that finding are credible, by avoiding incorrect results and false
positives. Consistency can be checked using a GRIM (granularity-related
inconsistency of means) test, self-checks can be done to see whether reported
descriptive statistics are possible and statcheck can be used to check that
p-values are consistent with summary statistics and degrees of freedom. Making data
publicly available, documenting them with metadata, a data dictionary or a
codebook, and sharing data processing and analysis code ensures reproducibility
of results. Dynamic reporting integrates code within a paper. False positives
can be avoided through blind analysis and through preregistration of the hypotheses
and analyses before data inspection or data collection. This plan is then published
in an non-editable and time-stamped way. A multiverse analysis repeats the
planned analysis across all the datasets that can be reasonably constructed from
the raw dataset.