A small subset of explainability techniques developed initially for image\nrecognition models has recently been applied for interpretability of 3D\nConvolutional Neural Network models in activity recognition tasks. Much like\nthe models themselves, the techniques require little or no modification to be\ncompatible with 3D inputs. However, these explanation techniques regard spatial\nand temporal information jointly. Therefore, using such explanation techniques,\na user cannot explicitly distinguish the role of motion in a 3D model's\ndecision. In fact, it has been shown that these models do not appropriately\nfactor motion information into their decision. We propose a selective relevance\nmethod for adapting the 2D explanation techniques to provide motion-specific\nexplanations, better aligning them with the human understanding of motion as\nconceptually separate from static spatial features. We demonstrate the utility\nof our method in conjunction with several widely-used 2D explanation methods,\nand show that it improves explanation selectivity for motion. Our results show\nthat the selective relevance method can not only provide insight on the role\nplayed by motion in the model's decision -- in effect, revealing and\nquantifying the model's spatial bias -- but the method also simplifies the\nresulting explanations for human consumption.\n