|
Abstract:
|
In the field of functional magnetic resonance imaging, it is crucial yet challenging to quantify the reproducibility of the generated data because of its high dimensional nature and the usually complex measuring and preprocessing procedures. Novel data reproducibility measures have been brought up in the context where a set of subjects are measured twice or for multiple times, including fingerprinting, rank sums, discriminability, and various generalizations of intra-class correlation. However, the relations between and the best practices among these measures remain largely unknown. In this manuscript, we systematically analyze the most natural reproducibility statistics associated with different statistical models. We show that the rank sum statistic is deterministically linked to an estimator of discriminability. We theoretically prove the relation between discriminability and intra-class correlation under mixed effect models, with univariate or multi-variate measurements. Powers of permutation tests derived from these measures are compared numerically under Gaussian and non-Gaussian settings, and with simulated batch effects. Recommendation is given for each setting accordingly.
|
