Candidate models explaining variation in senescence.

We used the Weibull model [28] to estimate senescence rates. The Weibull model has two parameters that define the hazard function: ��, which represents the shape of the hazard function, and ��, which represents the magnitude of the hazard given the shape of the function. We also included a day of entry parameter, b, because previous work has shown that early breeding salmon senesce slower than late breeding salmon [16]. Listed is the general model structure, the negative log-likelihood (NLL), the number of parameters (No. parameters), Akaike's Information Criterion (AIC), and the ��AIC (relative to the best model, model VII). See Table S1 of the Electronic Supporting Information for explicit formulae for each of the above models.