Table 2.

Alternative hypotheses that may account either partially or fully for the observed correlation between grazing (g) and coral recruitment (r)

Alternative hypothesisTestsResult
(i) Because recruitment in the Caribbean is dominated by brooding species whose larvae have short planktonic phases (50), recruitment density may be driven by the local cover of brooding corals. A positive correlation between grazing intensity and recruitment could then be observed because, coincidentally, grazing intensity is also correlated to the adult cover of brooders (and spawners) because such areas have higher rugosity and availability of shelter. r(r,c) = 0.07, P = 0.84, 1 − ß = 0.07, n = 9Reject
r(c,g) = 0.31, P = 0.37, 1 − ß = 0.26, n = 9
(ii) Corals recruit in cryptic areas and recruitment density is positively correlated to rugosity. This is particularly true of settling corals, although the degree to which it holds for larger recruits that have survived early postsettlement processes is unclear. Coincidentally, grazing is also correlated to rugosity of the habitat. r(r,rg) = 0.21, P = 0.59, 1 − ß = 0.54, n = 9Reject
r(rg,g) = 0.4, P = 0.28, 1 − ß = 0.18, n = 9
(iii) Increased levels of predation in reserves reduce the biomass and density of territorial damselfish (Pomacentridae). Reduced numbers of damselfish allow enhanced survival of coral recruits because there are fewer territories (56) and/or reduce the amount of macroalgae in territories. In this scenario, the correlation between parrotfish grazing and either coral recruitment or macroalgal cover arises because parrotfish grazing responds to the reserve in the same way as predator abundance, yet it is the latter that influences recruitment and macroalgal cover because of predator impacts on damselfishes. r(g,d) = 0.31*, P = 0.43, 1 − ß = 0.33, n = 9Reject
r(r,d) = −0.06*, P = 0.89, 1 − ß = 0.07, n = 9
Overall test of hypotheses i–iii seeking evidence for conditional dependence of main effect r(r,g) on other factors rp (r,g·c,d,rg) = 0.98, P = 0.0002, n = 9. Reject spurious correlation. Also, because strongest nonsignificant relationship is r(rg,g), rp (r,g·rg) = 0.87, P = 0.0054. Reject.
(iv) Elevated coral recruitment in no-take marine reserve (where fish grazing is greater) is due to exceptionally high larval supply. Although this would not disprove the main hypothesis, because elevated grazing and larval supply could enhance recruitment together, it could constitute an alternative explanation if larval supply were the only factor involved in determining recruitment. Tested using Lagrangian simulation model of larval dispersal.GLM no. coral larvae P time < 0.001, P site = 0.12DP = 3.9Reject
  • Other variables are denoted c (coral cover), rg (rugosity), and d (pomacentrid biomass). r, Pearson correlation coefficient; rp , partial correlation coefficient with conditional variables separated by ·; P, probability that coefficient = 0; 1 − ß, power; GLM, generalized linear model with time (serial autocorrelation, measured by using ACF (empirical autocorrelation function), removed using 5-day aggregate) and site as fixed factors, quasipoisson errors, and a log link (55); DP, dispersion parameter for quasipoisson family.

  • *Similar result using density of adult damselfishes.