Patterns of diversity varied across the 15 studies (Figure 3). Many studies showed little diversity change, particularly studies of long-term forest plots (Bradford et al. 2014). In these studies, total diversity change is typically zero and the contribution of each mechanism is negligible. Other studies are far more dynamic such as communities experiencing disturbance regimes (Webb and Scanga 2001, Thorn et al. 2016) or regeneration (Venturoli et al. 2011) experienced strong rarity shifts. In these studies, all mechanisms could make a substantial contribution to diversity change. Ernest et al. (2009) studies seasonal flowering xeric shrubland communities in the deserts of Arizona each summer and winter. Immigration had an unusually strong strongly influenced diversity. Immigration increased diversity strongly in some plots and decreased diversity strongly in others. The high immigration rate reflected the high turnover in species, with the dominant species frequently being replaced, largely due to the dramatic fluctuations in precipitation (Baldwin et al. 2002). ­A similar experiment examines flowering plant communities in the desert of Arizona within plots where small mammal have been excluded (Ernest et al. 2009). Immigration tends to increase diversity, rarity shifts tend to decrease diversity and selection appeared to take a wide range of positive and negative values. In this community, the most abundant species is the Creosote bush, Larrea tridentata . This species is perennial and its abundance changes little over the course of the study. Many of the rarer species are annuals whose abundances fluctuate dramatically from one season to another. Figure 3: Partitions of Shannon diversity change across all plots and time periods for each of the 15 studies. Diversity change (purple) is partitioned into selection (red), immigration (green) and rarity shifts (blue) as is described in Equation 4. The dotted line represents zero diversity change. To improve readability, we set the x-axis from -.5 to .5, excluding (3.7%) of values. See supporting information for more details and Gini-Simpsons diversity comparison.
Generalized Additive Models indicate that there is a non-linear relationship between the strength of rarity shifts related and the strength of other mechanisms. Rarity shifts were close to zero when selection was zero, but tended to become negative when selection was non-zero. Rarity shifts were close to zero when immigration was close to zero and was slightly positive when immigration was non-zero (Figure 4). Each term in each model required a relatively complex spline with many effective degrees of freedom (selection edf = 7.8, immigration edf = 8.8). Only a modest amount of variation was explained by the model (R2adjusted = 0.345). Supporting information shows similar trends in Gini-Simpsons diversity.