Search

Multimedia data from two probability-based exposure studies were investigated in terms of how censoring of nondetects affected estimation of population parameters and associations. Appropriate methods for handling censored below-detection-limit(BDL)values in this context were unclear since sampling weights were involved and since bivariate associations/measures were of interest. Both simple substitution(e.g., using 1/2 or 2/3 of the detection limit(DL)for BDL values)and truncation-based strategies were investigated by creating some artificial DLs and comparing resultant estimates with the original studies'uncensored results. The substitution methods generally outperformed the truncation methods, with the(2/3)DL substitution generally performing best.

Multimedia data from two probability-based exposure studies were investigated in terms of how missing data and measurement-error imprecision affected estimation of population parameters and associations. Missing data resulted mainly from individuals'refusing to participate in certain measurement activities, rather than from field or laboratory problems; it suggests that future studies should focus on methods for maximizing participation rates. Measurement error variances computed from duplicate-sample data were small relative to the inherent variation in the populations; consequently, adjustments in nonparametric percentile estimates to account for measurement imprecision were small. Methods of adjustment based on lognormality assumptions, however, appeared to perform poorly.