|
Abstract:
|
For a given research question of interest, there is usually a multiplicity of possible analysis strategies as many different analytical pathways are acceptable. If this flexibility is combined with selective reporting, it can lead to an increase in false positive results, inflated effect sizes, overoptimistic measures of predictive performance, and ultimately, to research findings that cannot be replicated in subsequent studies. The flexibility of analysis strategies can be explained by six sources of uncertainty that are ubiquitous across a broad range of scientific disciplines: sampling and measurement uncertainty, which are not under the control of the researcher, and model, parameter, data pre-processing and method uncertainty, which lead to a multiplicity of possible analysis strategies. In this talk, I will present an overview of solutions to address the multiplicity of analysis strategies that have been proposed across disciplines to generate findings more likely to be replicated in future studies and more broadly about ways to improve the modelling and communication of evidence and uncertainty in science.
|
