Abstract
This paper presents an approach for incorporating reliability on initial performance prediction models developed from as little as two time series predictors. It employs a novel methodology to provide apparent ages as surrogate of condition and, in addition, applies multilevel Bayesian regression to calibrate mechanistic empirical models to local conditions. This paper develops an international roughness index deterministic performance model for the Costa Rica road network and, further shows the procedure for obtaining a probabilistic multilevel Bayesian model which includes distributions of the mechanistic parameters and confidence intervals for the predicted performance. Bayesian statistics are also deployed for calibrating pavement strength coefficients to local observations.