Katz centrality is a popular network centrality measure. It takes a (weighted) count of all walks starting at each node, with an additional damping factor of $\alpha$ that tunes the influence of walks as lengths increase. We introduce a tool to compare different centrality measures in terms of their node rankings, which takes into account that a relative ranking of two nodes by a centrality measure is unreliable if their scores are within a margin of error of one another. We employ this tool to understand the effect of the $\alpha$-parameter on the lengths of walks that significantly affect the ranking of nodes. In particular, we find an upper bound on the lengths of the walks that determine the node ranking up to this margin of error. If an application imposes a realistic bound on possible walk lengths, this set of tools may be helpful to determine a suitable value for $\alpha$. We show the effect of $\alpha$ on rankings when applied to the Susceptibility Inference Network, which contains subject matter expert informed data that represents the probabilities of medical conditions progressing from one to another. This network is part of the Medical Extensible Dynamic Probabilistic Risk Assessment Tool, developed by NASA, an event-based risk modeling tool that assesses human health and medical risk during space exploration missions.
