Separating Measurement Error and Signal in Environmental Data: Use of Replicates to Address Uncertainty
Separating Measurement Error and Signal in Environmental Data: Use of Replicates to Address Uncertainty
Measurement uncertainty has long been a concern of scientists, engineers, and others in characterizing and interpreting environmental and toxicological measurements. This work incorporates replicates in an analysis of variance (ANOVA) model to determine how much of the estimated variance is attributable to measurement error versus signal, and it provides improved estimation for confidence intervals, enhanced interpretability of results, and increased understanding of measurement uncertainty. We use a simulation study and case studies to compare three statistical analysis approaches: a Naïve approach that ignores replicates, a Hybrid approach that treats replicates as independent samples, and a Measurement Error Model (MEM) approach that uses the ANOVA model incorporating replicates. The simulation study assesses effects of sample size and levels of replication and signal and measurement error variance on estimates from the three statistical approaches. The case studies analyze data for normally distributed arsenic levels in tap water and log-normally distributed lead concentrations in tire crumb rubber and calculate MEM confidence intervals for the true, latent signal mean and latent signal geometric mean (i.e., with measurement error removed), respectively. The MEM approach presented here applies established statistical theory to address and reduce measuring-induced uncertainty and inform sampling designs for optimizing replicate sample collection.