Causal networks clarify productivity–richness interrelations, bivariate plots do not

摘要: Summary Perhaps no other pair of variables in ecology has generated as much discussion species richness and ecosystem productivity, illustrated by the reactions Pierce (2013) others to Adler et al.'s (2011) report that empirical patterns are weak inconsistent. et al. argued we need move beyond a focus on simplistic bivariate relationships test mechanistic, multivariate causal hypotheses. We feel continuing debate over productivity–richness (PRRs) provides focused context for illustrating fundamental difficulties using gain scientific understanding. Pierce disputes conclusion ‘weak variable’. He argues, instead, data actually strong and, further, failure adhere humped-back model (HBM; sensu Grime 1979) threatens scientists' ability advise conservationists. Here, show Pierce's reanalyses invalid, statistically significant boundary relations difficult detect when proper methods used his advice neither advances understanding nor quantitative forecasts needed decision makers. We begin examining Grimes' HBM through lens networks. first translate ideas contained into diagram, which shows explicitly how multiple processes hypothesized control biomass production their interrelationship. then evaluate diagram structural equation modelling example from published study meadows Finland. Formal analysis rejects literal translation reveals additional at work. This exercise practice abstracting systems networks (i) clarifies possible hypotheses, (ii) permits explicit testing (iii) more powerful useful predictions. Building Finnish meadow example, contrast utility plots compared with models investigating underlying processes. Simulations illustrate fallibility means supporting one theory another, while based can quantify sensitivity diversity both management natural constraints. A key piece critique conclusions relies upper regression he claims reveal between original data. demonstrate this technique associations purely random is invalid because it depends uniform distribution. instead perform quantile site-level summaries plot-level (using mixed-model regression). Using variety nonlinear curve-fitting approaches, were unable humped-shape reiterate variable. We urge ecologists consider advance our analysis. Further, emphasize network conceptualization also provide meaningful guidance conservation than perspective. Measuring only two does not permit evaluation complex resolve debates about mechanisms.