Metrics for quantifying and comparing observer error across vegetation studies

Observer error, a type of nonsampling error, is pervasive in vegetation sampling and often of a consequential magnitude. Observer error rates should be reported along with published studies, although there currently exists no standardized, easily comparable format. Here we describe five key metrics of observer error (i.e. imprecision between observers), how they are calculated, and how they could be reported and interpreted. Three metrics apply to species composition: pseudoturnover, observer bias in species richness, and underestimation of true species richness. Two metrics—cover agreement and observer bias in cover estimation—apply to categorical cover estimation. All metrics are simple to determine, could be calculated from virtually any multispecies sampling effort using two or more observers, and are easily compared with other studies. The metrics are all reported as percentages, allowing for relative comparisons among studies with greatly differing species diversities. We also describe how to decompose the amount of error in species composition and cover estimation into random and bias components. Such decomposition is useful in determining whether additional training may be necessary for some observers. Two of the five metrics—pseudoturnover and cover agreement—have been quantified in previous studies, and we compile a list of published rates of pseudoturnover within general habitat types, and published cover agreement categories, for comparison with future studies. Finally, we provide an example by calculating the observer error metrics for a real data set collected by three different observers.