Graph Convolutional Networks (GCNs) excel at learning and extracting meaningful features from data represented by graphs. However, their decision-making processes are complex and not readily interpretable. Post-hoc explanation methods, such as saliency map generators (SMG), like Grad-CAM or Grad-CAM++, address this issue by highlighting areas of input data that influence the model decisions. Although initially applied to image classification, SMG can be adapted to interpret the roles of nodes and edges in graph-based regression or classification tasks. This paper introduces Single Step Metrics (STM), novel metrics designed to evaluate the performance of SMG in graph regression tasks, offering a new approach for quantitatively assessing the interpretability of GCN models. By applying these metrics, we demonstrate that the STM results align with the insights provided by SMG. Specifically, we compare the performance of Grad-CAM and Grad-CAM++ across three chemical datasets, proving that STM can effectively differentiate between SMG methods and identify the one better suited for a given problem. Our findings indicate that both SMG and STM consistently show that Grad-CAM is more suited for analyzing graph-based chemical data.