This paper examines the efficacy of a multivariate statistical modeling approach to analyze public transit bus driver distraction data collected through a self-administered driver survey. The distracting activities were classified into four risk zones according to distraction risk indices derived from distracting ratings, distracting durations, and driver perception of risks. A multinomial logistic regression model was formulated for highly-risky distracting activities using levels of distraction as the categorical dependent variable and correlating it with categorical and continuous independent variables responsible for the distraction. Results revealed that the common sources of distraction were due to passenger-related activities, which match two-thirds of simulated validation outputs. On-site route observations and discussions with transit staff revealed mixed results. The model could be used to identify drivers at highest distraction risk from their demographic backgrounds as well as driving schedules. The transit agency can use the results to implement relevant policies and training programs to mitigate distraction and improve transit performance.