Predicting preliminary engineering costs for highway projects

Preliminary engineering (PE) for a highway project encompasses two efforts: planning to minimize the physical, social, and human environmental impacts of projects and engineering design to deliver the best alternative. State transportation agencies strive to manage these efforts efficiently, seeking to maximize the utilization of limited funding and workforce productivity. Managers need a feasible PE budget early in project development. The results reported herein will enable engineers and managers to develop PE budgets during the preconstruction phase of highway project development. Typically, transportation managers establish a project's PE budget using a percentage of estimated project construction costs disregarding other project-specific parameters. The commonly accepted rule of thumb for PE costs is 10% of estimated construction costs. This research effort sought to improved PE cost estimation methods by investigating available historical data supporting statistical analyses, developing predictive regression models forecasting projects' PE costs, and assessing model performance through validation. Cost data were acquired for 461 bridge projects and 188 roadway projects let for construction between 1999 and 2009 in North Carolina. Separate analyses were performed on bridges and roadways. Both project types are included in North Carolina's State Transportation Improvement Program. Many variables applicable for bridges were not applicable or available for roadways. Analysis of the roadway data yielded a mean ratio of PE cost to estimated construction cost (the PE cost ratio) of 11.7%. Comparatively, the bridge projects exhibited a mean PE cost ratio of 27.8%, approximately 2.4 times greater than the roadway mean. Using multiple linear regression and hierarchical linear modeling, we developed prediction models to forecast the PE cost ratio of future bridge and roadway projects. Before model development began, we randomly selected a portion of the data for validation. Regression modeling used the remainder of the data. Validation of the bridge model (using 70 projects) resulted in a mean absolute percentage error (MAPE) of 43%. The bridge model utilized eight variables, four numerical and four categorical, with interactions. When validating on 38 roadway projects, the roadway model yielded an average absolute error (AAE) of ± 6.91%. Considering the roadway database exhibited a mean PE cost ratio of 11.7%, this AAE represents a 59% relative error. Comparatively, predictive regression modeling did yield better PE cost estimates than using the commonly accepted rule of thumb, or any single parameter estimator. However, the relative error was high, in the 40% to 60% range. Example calculations using the developed models (bridge and roadway) are presented. Improvements to agency preconstruction cost accounting and tracking processes would support better modeling efforts. Lack of consistent historical data influenced our model development. Others pursing PE cost analyses should thoroughly evaluate and assess data quality before initiating similar modeling efforts.