While motorized travel provides many benefits, it can also do serious harm in the form of road-related collisions. The problem affects millions of human lives and costs billions of dollars in economic and social impacts every year. The problem could be addressed thorough several approaches with engineering initiatives being recognized as the most sustainable and cost effective. However, the success of the engineering approaches in reducing collision occurrences hinges upon the existence of reliable methods that provide accurate estimates of road safety. Currently, Safety Performance Functions (SPFs) are considered by many as the main tool in estimating the safety levels associated with different road entities. Therefore, the research in this thesis focuses on addressing key issues related to the development of SPFs for i) collision data analysis and ii) collision intervention analysisSome of the key issues addressed include: 1) adding spatial effects to SPFs thereby recognizing the evident spatial nature of road collisions, 2) fitting hierarchical models to allow inference to be made on more than one level, 3) recognizing the multivariate nature of collisions as most data are available by collision type or severity and modeling the data as such, 4) identifying and accounting for outliers in the development of SPFs, 5) developing a novel evaluation methodology to estimate the effectiveness of safety countermeasures when subject to data limitations, and 6) compare different tools for investigating the safety change in treated sites due to the implementation of safety countermeasures. The applications of the various models have been demonstrated using several collision datasets and/or safety programs. The results provide strong evidence for (i) incorporating spatial effects in SPFs, (ii) clustering road segments or intersections into homogeneous groups (e.g., corridors, zones, districts, municipalities, etc.) and incorporating random cluster parameters in SPFs, (iii) developing robust multivariate models with multiple covariates for modeling collisions by severity and/or type concurrently, and (iv) the effectiveness of the proposed full Bayes safety assessment methods that account for several theoretical and practical issues concurrently. In addition to the improvement in goodness of fit, the proposed models have also improved inference and precision of expected collision frequency.