This amazing study implements at a landscape scale the type of experimental design more commonly observed in microcosms or old-fields. It is awesome!
An important difference between this experiment and those of microcosms is that microcosms run for many (10-100+) generations of the organisms, whereas this experiment ran (at the time of publication) for probably less than one generation for most of the organisms (recall generation = average age at parenthood). In that sense, it is less likely to confound evolutionary and ecological processes, so in that sense it is analogous to a pulse experiment.
The stats nerd appreciated Fig. 2C (y=Difference in richness between connected and unconnected, x=year). I wish that they reported more results in the text in terms of linear coefficients or effect sizes or actual differences, instead of the F-stats and P values. I would rather see them state biologically meaningful values, such as "The species richness of native species increased over time in connected (ave=4 spp/y +/- 1 spp) but not unconnected plots (ave=0 spp/y +/- 2 spp)."
I know it is a Science paper, but I wanted a lot more info in supplementary docs about which species were where. I am not sure why I want to know -- I should come up with hypotheses before I see the species list.
I could not figure out how they did their Chi-squared analysis (Table S5) which tests differences from expected frequencies of rare and common species in connected vs. unconnected patches. My guess is they did it right, but if not, it wouldn't be the first time if someone published bad stats in a prominent journal.
Adding a connector is simply increasing island area by doubling the size of the island from one patch to two patches. It need not be true that increasing island size increases average &alpha diversity of small samples on an island, but it would be true for a well-mixed island. So -- is this just the effect of area?