Bootstrapping for Characterizing the Effect of Uncertainty in Sample Statistics for Braced Excavations

Abstract

A simple procedure for assessing the probability of serviceability failure in a braced excavation involving bootstrapping to characterize the effect of uncertainty in sample statistics is presented. Here, the failure is defined when an excavation system’s response in terms of the maximum wall deflection or ground settlement that exceeds the limiting value specified by the client or in an applicable code. The analysis for the probability of failure (or probability of exceedance) necessitates an evaluation of the means and SDs of critical soil parameters. In geotechnical practice, these means and SDs are often estimated from a very limited data set, which can lead to uncertainty in the derived sample statistics. Thus, in this study bootstrapping is used to characterize the uncertainty or variation of sample statistics and its effect on the failure probability. Through the bootstrapping analysis, the probability of exceedance can be presented as a confidence interval instead of a single, fixed probability. The information gained should enable the engineer to make a more rational assessment of the risk of serviceability failure in a braced excavation. The study points to the potential of the bootstrap method in coping with the problem of having to evaluate failure probability with uncertain sample statistics.