Post-stroke behavioural symptoms often correlate and systematically co-occur with each other, either because they share cognitive processes, or because their neural correlates are often damaged together. Thus, neuropsychological symptoms often share variance. Many previous lesion-behaviour mapping studies aimed to methodologically consider this shared variance between neuropsychological variables. A first group of studies controlled the behavioural target variable for the variance explained by one or multiple other variables to obtain a more precise mapping of the target variable. A second group of studies focused on the shared variance of multiple variables itself with the aim to map neural correlates of cognitive processes that are shared between the original variables. In the present study, we tested the validity of these methods by using real lesion data and both real and simulated data sets. We show that the variance that is shared between post-stroke behavioural variables is ambiguous, and that mapping procedures that consider this variance are prone to biases and artefacts. We discuss under which conditions such procedures could still be used and what alternative approaches exist.
Keywords: VLSM; lesion-deficit inference; nuisance regression; principal component analysis; statistical parametric mapping.
© 2019 The Authors. Human Brain Mapping published by Wiley Periodicals, Inc.