Table. 1.

Qualitative differences between neural score methods

Neural score	Neural score description
Informational network	- Univariate voxel threshold - Multivariate score computation - Categorical a priori representational model
Searchlight RSA	- Multivariate voxel threshold - Multivariate score computation - High-dimensional a priori representational model
Whole-brain univariate	- Univariate voxel threshold - Univariate score computation - No a priori representational model

Each neural score derivation involved voxel selection and score computation steps. The informational network score involved a combination of univariate voxel thresholding with multivariate score computation, whereas the other two scores were derived using either entirely univariate or entirely multivariate methods. In addition, the informational network and RSA neural scores incorporated a priori models, whereas the univariate neural score did not. The informational network score incorporated a dimensionality-reduced version of the expert mechanical similarity model by using a category-based SVM classifier, whereas the RSA neural score incorporated the full 24-dimensional expert mechanical similarity model. No neural score method involved any factors with a direct relationship to the concept knowledge measures they were designed to predict