Nonparametric Approach to Managing Materials Quality

Abstract

The normal distribution was tested for its goodness of fit to quality control data from 25 California Department of Transportation projects. The binder content and relative compaction of asphalt concrete were studied. In a significant number of cases, the assumption of normality was rejected. Closer examination of the observed distributions revealed instances of skewed, truncated, and bimodal distributions. Indications of nonnormality were found for a third of the asphalt content cases and nearly half of the relative compaction cases. An alternative nonparametric procedure for estimation of the percentage of material outside specification limits (POL) is proposed. Comparisons with the current methodology, which assumes that the data are normally distributed, indicated that somewhat lower, but arguably more accurate, estimates of POL are obtained by the nonparametric approach. Issues of accuracy and precision were addressed through a simulation study. The nonparametric method offers considerably greater statistical accuracy and power than the parametric approach when the data are not distributed normally and a performance comparable to that of the parametric approach when they are.