Abstract: Observer error is a ubiquitous component of vegetation community sampling, yet it is frequently ignored. We conducted an analysis of observer error in herbaceous vegetation sampling by three observers over five years (2018-2022). A total of 94 plots were double sampled at six prairie and five woodland understory research areas in the midwestern U.S. Sampling was conducted at sites containing two sets of five plots each; plots were 10 m2 in size. Each of two observers recorded the species present and estimated abundances based on foliar cover within seven categories for all five plots in one set. One of the five plots in each set was then chosen at random and was sampled by the other observer. The goal of this analysis was to quantify interobserver error and its components over larger spatial and temporal dimensions than is usually done. Pseudoturnover—the apparent change in species composition due to observer error—ranged from 20.2 to 22.1% at prairie sites, and from 16.8 to 28.6% at woodland understory sites. Overall, there was no evidence that observer error declined over the five-year interval, despite knowledge of error rates by observers and annual trainings designed to reduce such error. Most pseudoturnover was due to overlooking species rather than misidentifications: Overlooking error was 16.0 to 21.4% across all sites, whereas misidentification error was only 0 to 4.3%. Overlooking errors occurred more frequently with less abundant species. There were significant differences between observer pairs in number of species recorded. Interestingly, the least experienced observer consistently recorded more species. Overlooking error due to interobserver bias (i.e., apparent inherent differences between observers) was about one-quarter to one-half (depending on observer pair) of the total overlooking rate. Rates of estimation error, which occurs when abundances are not accurately estimated, were similar, albeit more variable, to those of pseudoturnover. Observers recorded different categories 21.5 to 28.7% of the time at the prairie sites, and 14.4 to 43.1% of the time at woodland understory sites. When cover categories didn’t match, they were almost always (94 to 100% of the time) off by only one category. Unlike pseudoturnover, estimation errors occurred more frequently with more abundant species. Both pseudoturnover and estimation error were more variable at woodland understory sites than at prairie sites. Overall, observers generally differed by ~20 to 30% in estimates of both species composition and abundance. Observer error should always be quantified and reported in vegetation community studies.
