We often quantify the effect of environment on changes in community species composition as an explained variation (R² or adjusted R²) by constrained ordination methods. Here, we show that absolute values of explained variation cannot be compared among datasets with samples of different completeness. In our context, we define the sample completeness as the proportion of species theoretically occurring in the species pool of the sample which has been truly sampled. Community with high species pool surveyed by samples with few individuals will be recorded with lower completeness than the community with smaller species pool. At the dataset level, this incompleteness transfers into a higher proportion of rare species, which in turn decreases the absolute value of R² in constrained ordination. We used simulated and real vegetation data to show this; while keeping the effect of environment fixed, we vary the size of species pool and the number of individuals sampled within each sample, to show that R² increases with the completeness of the survey. Our main message is that absolute values of R² explained by environmental variables should not be compared between datasets unless we can guarantee that the completeness of sampling is comparable.