Measuring cost and time overruns: median-based metrics, reform effect, and simultaneous estimation
研究了挪威公路项目成本和工期超支的测量方法，发现中位数百分比误差比均值更准确，并评估了政府改革的效果，同时用三阶段最小二乘法纠正了成本与工期超支之间的内生性偏差。
This paper addresses three key gaps in transportation research on cost and time overruns: (1) the overuse of Mean Percentage Error (MPE), despite the Median Percentage Error (MdnPE) being more appropriate for skewed data; (2) limited evaluation of government interventions aimed at reducing overruns; and (3) the failure to account for the interdependence between cost and time overruns, leading to biased estimates. Using a dataset of 2,228 Norwegian road projects (1993–2016), we assess overruns using both MdnPE and MPE, evaluate the effects of major organizational reforms, and apply a Three-Stage Least Squares (3SLS) estimation to correct for endogeneity between cost and time overruns. Unlike some studies, we rely solely on statistically observable factors and avoid subjective explanations such as the planning fallacy or strategic misrepresentation. Results show a median cost overrun of 4% (mean: 12%) and a median time overrun of 0% (mean: 18%), confirming a skewed data distribution. This demonstrates how MPE can exaggerate typical performance, while MdnPE offers a more representative measure of central tendency. Still, the choice between MPE and MdnPE should depend on the analytical objective—whether the goal is to highlight extreme cases or typical project performance. We find that government reforms, particularly the introduction of full procurement competition and a Quality Assurance (QA) regime, significantly reduced overruns. However, results for time overruns vary by method: the Kruskal–Wallis test shows lower median time overruns during the QA and competition periods, while the 3SLS model—controlling for other project factors—reveals that the transition to full competition was associated with increased time overruns. This suggests that improvements in medians may reflect changing project composition rather than reform effects alone. This study underscores the need to consider how overruns are evaluated and encourages the use of empirical evidence over speculative explanations like the planning fallacy.