A Predictive Modeling Approach to Estimating Seismic Retrofit Costs.

This paper presents a methodology for estimating seismic retrofit costs from historical data. In particular, historical retrofit cost data from FEMA 156 is used to build a generalized linear model (GLM) to predict retrofit costs as a function of building characteristics. While not as accurate as an engineering professional's estimate, this methodology is easy to apply to generate quick estimates and is especially useful for decision makers with large building portfolios. Moreover, the predictive modeling approach provides a measure of uncertainty in terms of prediction error. The paper uses prediction error to compare different modeling choices, including the choice of distribution for costs. Finally, the proposed retrofit cost model is implemented to estimate the cost to retrofit a portfolio of federal buildings. The application illustrates how the choice of distribution affects cost estimates.