This research contributes to improved risk analysis of performance measure forecasts in road safety engineering by designing and applying a method to characterize uncertainty associated with forecast input data in cases where input uncertainty is not known. The research applies this method to quantify uncertainty in three categories of inputs used in risk analysis of performance measure forecasts in road safety engineering: (1) estimates of pedestrian exposure to collision risk; (2) estimates of vehicular exposure to collision risk; and (3) estimates of engineering economics parameters that assign valuations to mortality risk reductions based on individual willingness to pay. The common methods used in each of these categories are repeated comparisons of input ground truth to input estimations, the use of simulation approaches (e.g. the simulation of short-term counts by sampling permanent count data), and the use of non-parametric techniques to characterize input uncertainty. Some highlights of quantified input uncertainty levels include: (1) when obtaining pedestrian risk exposure estimates at a site in Winnipeg, MB by expanding two-hour short-term counts using the National Bicycle and Pedestrian Documentation Project method, 90% of errors are between 62% and 170%; (2) when obtaining estimates of vehicle exposure to collision risk by expanding two 48-hour counts using the individual permanent counter method for Manitoba highways, 92 % of errors are between 9.5% and 10.8%; and (3) when applying an income-disaggregated transfer function to estimate value of a statistical life for road safety in developing countries, 90% of errors are between 53% and 54%. The results provide further detail on the structure of these input uncertainties. Analytic and computational capabilities in forecasting and risk analysis have advanced beyond our understanding of corresponding input uncertainty levels; this research closes some of this gap and enables better risk analysis of performance measure forecasts in road safety engineering.