Quantitative relationships between TCE and 1,4-dioxane levels were evaluated for installations where significant
categorical association was observed. Trend analysis was carried out using quantile regression because heteroscedasticity
was observed (discussed below) as described by Koenker and Bassett (1982); heteroscedasticity results in biased
statistical inference for conventional regression methods predicated on the central tendency (i.e., the mean) of the
relationship with constant variance assumption (Montgomery et al. 2001). By evaluating the trend at various distributional levels (i.e., quantiles) of the relationship (as opposed to the mean or ‘‘central tendency’’), quantitative trends can be
tested with reliable inference when nonconstant variance is observed (Koenker and Bassett 1982; Buchinsky 1998).
Quantile regression was performed using Proc Quantreg in SAS Version 9.2 for Windows (SAS Institute 2008), and
results were considered significant at p0.05.