Forecasting State-Level Construction Labor Earnings for Enhanced Project Cost Control: An Econometric and Deep-Learning Analysis of the Leading Economic Indicators

As a major input to several work packages, the labor element constitutes a critical component for successful performance of construction projects. Localized labor shortages, fundamental changes in prevailing wage laws, and historical shifts in the unionization rates of construction workers impair the adequate estimation of construction labor costs in diverse labor market dynamics. Meanwhile, existing studies have utilized national-level indicators to study the trends of construction labor costs, but the relationship between the multifaceted local economic factors and state-level construction labor costs remains understudied. This paper fills such a knowledge gap. A three-stage methodology is adopted: (1) data collection of state-level construction labor earnings and macroeconomic indicators as the target variable and predictors, respectively; (2) dimensionality reduction of the state-level macroeconomic indicators using principal component analysis (PCA), and identification of short- and long-term associations between the labor earnings and the macroeconomic indicators using Granger causality and the Johansen cointegration tests; and (3) prediction of the state-level labor earnings using vector error correction (VEC) and long short-term memory (LSTM) recurrent neural network models. The research methodology is demonstrated in the domain of 16 states in the US. Results indicate that in the Northeast states, labor earnings are linked to workforce size and participation rates. In the Midwest and South, inflation indicators consistently precede changes in construction worker earnings, whereas union representation is a reliable indicator of earnings in Illinois, Indiana, and West Virginia. The predictions revealed that multivariate LSTM captures the changes in labor earnings in the long-term forecasting horizons. This study can be replicated to augment the control of labor costs at other geographical domains. The developed multivariate prediction models provide owners and contractors with enhanced state-level estimating of construction labor costs, prescient cost planning in the tendering stage, and proactive control of schedules and budgets during execution.