Braun-Blanquet data in ANOVA designs: comparisons with percent cover and transformations using simulated data

The Braun-Blanquet (BB) cover-abundance scale is used to visually estimate community composition and species dominance. An 8-division variant was developed for benthic systems in the 1990s; the capacity to speed collection of seagrass coverage data led to its adoption by several large-scale monitoring programs in the USA. However, debate regarding how best to treat ordinal BB data in statistical analysis has stymied progress in the comparison of status and trends. Methods specific to ordinal data exist; however, they have generally been ignored in favor of transformation to percent cover or the use of BB categories as continuous data in parametric statistics and multivariate ordination. To quantify behavior of BB data in 1-way ANOVA, we conducted a series of data simulations using percent cover, BB scores and 3 metric-scale transformations as competing dependent variables in iterated 2-group contrasts. Simulations followed the design of the Fisheries Habitat Assessment Program (FHAP) and covered full ranges of within- and between-group variation. We empirically estimated Type I error and proportional deviance in effect size as measures of performance. Finally, we compared 6 yr of FHAP data to the simulations to identify scenarios likely to be encountered by seagrass ecologists. BB scores performed well as a proxy for continuous data and log-linear transformation allowed more precise effect size estimation. Our results highlight the need for high levels of replication in benthic sampling and provide empirical evidence for the statistical reliability of BB data in parametric analysis.