Cautions on the treatment of non-detect results for environmental forensics

ABSTRACT While censored data (i.e., containing non-detect [ND] results) may provide useful information in the context of risk assessment, their utility in chemical forensics is less clear. Substitution of NDs can create invasive patterns and may obscure the true forensic features in the underlying chemistry and sample compositions. The effect of substituting numeric values for NDs on the results of common chemical forensics multivariate analyses is illustrated using synthetic simulated data and laboratory measurements of soil samples from Colorado. The analyses of simulated and laboratory data demonstrate the impact of substituting values for NDs on multivariate analyses and indicate the potential for incorrect conclusions about the true forensic features within the data. Recommendations are provided for how to investigate and visualize the impact of NDs on chemical forensic multivariate analyses.