Subjectively sampled vegetation data: Don’t throw out the baby with the bath water

Abstract

Lájer (2007) reviewed the use of statistical methods for the analysis of vegetation data and concluded that their assumptions are often violated. Most importantly, the vegetation data are usually not a random sample of vegetation, and consequently the statistical tests should not be used at all. Although we appreciate many concerns raised in Lájer’s paper, we find his conclusions far-fetched and inappropriate for practical use. In large-scale observational studies, it is practically impossible to define the sampled statistical population in the way that would allow for truly random sampling. As we are unable to ensure that any part of the population can appear in our sample with the same probability, we must try to avoid interaction between any bias imposed by the sampling procedure with the hypotheses we test. The same stance can also be taken in the case of datasets extracted from the databases of phytosociological relevés, where a substantial bias was introduced by phytosociologists’ prejudice in selection of sampling locations. We can hardly use this kind of data when testing hypotheses concerning absolute values of species richness or discontinuities in community variation, but there are also many important research hypotheses that can be tested even with the help of this data source. In fact, such voluminous databases are the only available source of data for exploring research questions referring to large spatial or temporal scales. We must also reject Lájer’s suggestion that application of descriptive, exploratory statistical methods is less harmful for datasets with sampling bias than application of tests of statistical hypotheses. Finally, we stress the difference between observational and experimental studies, because the rigor of sampling design should be respected much more in the case of latter ones.