Construction Industry Productivity: Omitted Quality Characteristics in Construction Price Indices

Abstract

A fundamental challenge to the U.S. construction industry is the establishment of an accurate measure of the industry’s productivity. The inability to accurately measure construction inflation is cited as a primary reason why the U.S. Bureau of Labor Statistics does not maintain an official productivity measure for construction (Dacy 1965, Gordon 1968, Rosefielde and Mills 1979, Pieper 1990, Gullickson and Harper 2002). Macroeconomic studies have shown a decline in construction productivity (Stokes 1981, BRT 1983, Allen 1985, and Teicholtz 2000), while micro studies have shown an increase (Allmon et al 2000 and Goodrum et al 2002). One theory to explain these opposing results is that current measurements of construction inflation do not adequately consider the change in quality and amenities provided in today’s structures as compared to when the measurements were first established. An inflation index that is considered to be one of the industry’s major measurements of construction inflation is the U.S. Census Bureau New One-Family Houses under Construction Price Index. In this paper, the writers present initial results of a larger study to examine the current hedonic regression model used by the Census Bureau to estimate this price index, which is used to help measure a significant proportion of the industry’s real output. This study will compare the current hedonic model with a proposed model, which will include quality variables that are not part of the Census’ current hedonic model. Previous research suggests that omitted quality variables in the Census’ price index potentially leads to omitted variable bias which overestimates construction inflation leading to both an underestimate of construction industry output and productivity. However, previous research has not formalized necessary changes to the Census index to avoid this bias nor has the bias been actually measured in order to quantify its direction or magnitude. For this paper, price and quality data of new homes was collected from Multiple Listing Service data from Warren County, Kentucky for years 2002 through 2007, which was provided by the local Board of Realtors. The data was compiled to determine the frequency of additional quality characteristics currently not measured by the Census Bureau. By