Multiverse analysis in graph-based fMRI research

Improving replication rates in graph-based brain connectivity analysis requires that researchers disclose their degrees of freedom by making explicit the arbitrary but equally defensible choices in design, data processing, and statistical analysis. This approach, known as multiverse analysis, is gaining ground in cognitive neuroscience. However, knowledge of the full range of analytical options - the "garden of forking paths" - is essential to this approach. Through literature mining and expertise crowdsourcing, we have created a dynamic knowledge space of analytical decisions that can be used interactively by researchers. The space contains 61 different steps, 17 of which have controversial parameter choices as options. We show how this space can be used for the design and application of multiverse analysis in graph-based brain connectivity research, including an active learning-based sampling approach of potential pipelines to approximate an exhaustive multiverse analysis. We will discuss the implications and the next steps to be taken.