Short-term safety performance functions (SPFs) were proposed to achieve accurate and dynamic crash frequency predictions and bridge the gap between annual crash frequency prediction and real-time crash likelihood prediction. The proposed short-term SPFs consider the temporal variation in crashes and traffic characteristics. This study contributes to the literature by developing short-term SPFs at hourly aggregation levels for freeways that include high-occupancy vehicle (HOV) lanes using loop detector data from Arizona State, U.S. Variables that capture the short-term traffic turbulence were prepared and considered in the developed SPFs. Further, this study investigated the factors contributing to crash frequency using three different ways to represent the hourly traffic: annual average hourly traffic, annual average weekday hourly traffic (AAWDHT), and annual average weekday peak hour traffic (AAWDPT). The results indicated that the traffic volume variable was found to be significant in all the developed models. Further, the variables that represent the speed and occupancy differences between HOV lanes and general-purpose lanes were positively associated with crash frequency. This study proposed a series of variables that reflect the short-term traffic turbulence. The models comparison results showed an improvement in [Formula: see text] from 2.4% to 12.8% when including the proposed variables. Further, the results indicated that the Poisson-lognormal approach outperformed the basic negative binomial model in both AAWDHT, and AAWDPT models. Further, the AAWDPT model was found to have the best performance in relation to Akaike information criterion and [Formula: see text].