A methodology for estimating seismic retroft costs

Abstract

Presidential Executive Order 13717 (EO 13717), Establishing a Federal Earthquake Risk Management Standard, encourages federal agencies to "enhance resilience... [to] future earthquakes" by evaluating and retroftting existing federal buildings based on current existing building codes. However, while guidance on evaluation and retroft practices is readily available, a standard approach to estimating retroft costs does not exist. Moreover, the absence of easily obtainable estimates can make retrofts prohibitive for decision-makers.This paper develops a cost-estimating methodology for seismic retrofts that (1) captures the essential factors that drive seismic retroft costs, such as building construction and square footage; and (2) is reproducible using data available to decision-makers.The methodology builds on FEMA 156 and 157, Typical Costs for Seismic Rehabilitation of Existing Buildings, Volumes 1 and 2. A series of regression models is ft to the data used for the FEMA reports, with the models varying in the level of data required; e.g., a decision-maker may not have information on building construction for each asset in their inventory. Thus, the trade-off from estimating retroft costs subject to data limitations can be quantifed in terms of prediction error, providing decision-makers with a set of options for estimating costs together with a measure of predictive performance. We fnd that a simple model, in terms of data requirements, can deliver reliable predictions.

A predictive modeling approach to estimating seismic retrofit costs

“…FEMA 156 and FEMA 157 present the model, but do not test how well it performs in predicting retrofit costs. Fung et al (2017) show that a model with a different set of predictors, outperforms the FEMA 156 model in the sense of having a lower prediction error in predicting retrofit costs. For instance, Fung et al (2017) show that including an indicator for whether a building is deemed historic as an additional predictor lowers prediction error.…”

“…Elements of the predictive modeling methodology, inspired by FEMA 156 and FEMA 157 (FEMA, 1995), are developed by Fung et al (2017). FEMA 156 made two major contributions: (1) the collection of a reliable data set of retrofit costs and building characteristics; and (2) methods for estimating average retrofit costs, including a linear regression model.…”

“…Model assumptions: The predictive methodology developed by Fung et al (2017) is based on a standard linear regression model. This article uses a generalized linear model (GLM) and compares GLM to standard linear regression, measuring model performance in terms of prediction error.…”

“…Model selection and evaluation: Fung et al (2017) evaluate models using the “hold-out” method, where a subset of the data is held out when fitting (or “training”) a model. Thus, the hold-out method does not use all of the available data to train a model.…”

“…Thus, the hold-out method does not use all of the available data to train a model. Moreover, Fung et al (2017) focus on model evaluation, assessing how well a model performs, rather than model selection, choosing the best among several competing models. This article uses nested K - fold cross-validation to perform both model selection and model evaluation.…”