Incorporating measurement error in testing for changes in biodiversity

Abstract

Abstract The effect of measurement error on biodiversity metrics is problematic for conservation planning and testing ecological theories, but there are few methods for including it in statistical analyses. We present a framework for incorporating measurement error in testing for significant change in almost any type of biodiversity metric. We used repeat measurements of 20 × 20 m2 forest plots to quantify measurement error for species composition and mean stem diameter. With these data we illustrate generalizable methods for incorporating measurement error in testing for change in three types of metric—(1) species composition, (2) mean stem diameter, and (3) proportional richness of ungulate palatability groups—between surveys conducted 7 years apart on a national plot monitoring network. Turnover in species composition (Jaccard dissimilarity) from measurement error was considerable, with a mean of 36% (ranging from 23% to 52%), but observed turnover between surveys was significantly greater (for Jaccard and both its richness dependent and independent components) than expected from measurement error alone. Measurement error increased the standard deviation for diameter change by an average of 10%, and increased sample size needed to detect a pre‐defined change by an average of 21%, although in both cases variation across species was considerable (0%–50%, and −1%–110% respectively). The effects of measurement error on uncertainty for palatability groups were even more variable, ranging from 13% to 100% for standard deviation and from 26% to 299% for required sample size. Importantly, four of the eight species that yielded significant changes in mean diameter with observed data showed no significant shift in diameter when measurement error was included in paired t tests. Exclusion of measurement error can therefore alter conclusions drawn from monitoring networks, risking spurious claims of change. Our study demonstrates that measurement error can be incorporated in testing change for three numerically different (continuous, proportional and multivariate) classes of biodiversity metric at both the species and community level. This shows that it is possible to incorporate measurement error in assessing change for almost all biodiversity metrics.