These data include total anchovy and sardine biomass west and southeast of Cape Agulhas (sampled in November), and anchovy and sardine recruitment west of Cape Infanta (sampled in May). Seabird variables included % of the diet comprised of anchovy, % of the diet comprised of sardine, breeding success, and survival. African penguin data were collected at Dassen and Robben Islands, South Africa. Cape gannets data were collected at colonies on Malgas Island and in Lamberts Bay, South Africa.

These variables were entered into path analyses, run in Stata v. 11.1 (StataCorp). Path models were designed a priori with upwelling as the base predictor variable, fish biomass (both species) and anchovy recruitment as intermediate predictors, and a seabird metric as the response variable. Path analysis produces beta coefficients for each path segment of the model. To determine the dominant path of effect, the beta coefficients for each indirect path were multiplied together, then summed. This sum of the products of all of the indirect paths was compared to the beta coefficient of the direct path of effect. The greater value indicated if the dominant path was direct or indirect. If indirect, the path with the largest beta coefficient product was considered the dominant path of the model. Note that fish biomass, and anchovy recruitment data were ln-transformed prior to analysis. Anchovy and sardine biomass were lagged one year for analyses of penguin responses, to precede the penguin breeding season when those measurements were collected.