Regression-based prediction methods for adjusting construction cost estimates by project location

Construction cost estimates are fundamental to the success of a construction project. Several estimates are performed throughout a project lifecycle to make decisions on the project with stakeholders often relying on historical data to estimate costs of future construction activities. Location cost adjustment factors (LCAFs) are commonly used to adjust historically based estimates by project location with the selection of the LCAF being one important step of this process. Various datasets provide LCAF values for sampled locations, but, obviously, not all locations across North America are included. Therefore, spatial interpolation and prediction methods are needed to infer LCAF for un-sampled locations. The current industry practice is to select the value for a location using only one variable, namely the nearest linear-distance between two sites. Arguably, construction costs could be affected by other variables, including socio-economics. This research investigated relationships between a commonly used set of location adjustment factors, the City Cost Indexes (CCI) by RSMeans and other attributes, included in the ESRI Community Sourcebook. Regression-based prediction modeling was investigated to understand if it could be an appropriate way to model CCI as a function of multiple covariates. WEKA and ArcGIS packages were used to develop and test the prediction models. The prediction models did not outperform interpolation methods, as expected. In addition, among the two prediction models, the GIS-based regression (GISBR) model slightly outperformed the WEKA-based regression (WEKABR) model.