How should non-detect data be treated in statistical analysis of site data?

How should non-detect data be treated in statistical analysis of site data?

• A. Replacing non-detects with LOD/2
• B. Treat using censored data techniques or use statistical methods for censored data
• C. Exclude non-detects from analysis
• D. Replace non-detects with zero

Answer: (B)

Non-detect values are left-censored: the true concentration is somewhere below the detection limit but not known exactly. Treating them as exact values (using LOD/2 or zero) or excluding them discards information and can bias estimates of central tendency and variability. The proper approach is to analyze non-detect data with censored-data techniques, which explicitly account for the fact that some observations are only known to be below a threshold. This includes methods like maximum likelihood estimation under a censored distribution (often assuming a lognormal distribution for environmental concentrations), Tobit models, or nonparametric approaches akin to Kaplan-Meier estimators, depending on the analysis. Using these censored-data methods yields more reliable estimates of means, medians, and distribution shapes than simple substitution or exclusion.

Non-detect values are left-censored: the true concentration is somewhere below the detection limit but not known exactly. Treating them as exact values (using LOD/2 or zero) or excluding them discards information and can bias estimates of central tendency and variability. The proper approach is to analyze non-detect data with censored-data techniques, which explicitly account for the fact that some observations are only known to be below a threshold. This includes methods like maximum likelihood estimation under a censored distribution (often assuming a lognormal distribution for environmental concentrations), Tobit models, or nonparametric approaches akin to Kaplan-Meier estimators, depending on the analysis. Using these censored-data methods yields more reliable estimates of means, medians, and distribution shapes than simple substitution or exclusion.