Table 3.
Model selection table for candidate models of sample covariates and subsequent set of temperature metric and time lags
| Parameters | k | log likelihood | AICc | Delta | Weight |
|---|---|---|---|---|---|
| Sample covariate models | |||||
| Seasona | 6 | −233.33 | 478.69 | 0 | 0.49 |
| Season + speciesb | 7 | −233.09 | 480.23 | 1.54 | 0.23 |
| Season + agec | 7 | −233.33 | 480.70 | 2.01 | 0.18 |
| Season + age + species | 8 | −233.09 | 482.23 | 3.54 | 0.08 |
| Null (intercept only) | 1 | −242.50 | 487.01 | 8.32 | 0.01 |
| Species | 2 | −242.03 | 488.07 | 9.38 | <0.01 |
| Age | 2 | −242.50 | 489.00 | 10.31 | <0.01 |
| Age + species | 3 | −242.00 | 490.01 | 11.33 | <0.01 |
| Temperature lag additive modelsd | |||||
| Season + f(min 11‐d lag) | 8 | −236.28 | 488.62 | 0.00 | 0.45 |
| Season + f(mean 11‐d lag) | 8 | −238.16 | 492.38 | 3.76 | 0.07 |
| Season + f(min 10‐d lag) | 8 | −238.32 | 492.69 | 4.07 | 0.06 |
| Season + f(mean 11‐d lag) | 8 | −238.50 | 493.05 | 4.43 | 0.05 |
| Season + f(min 12‐d lag) | 8 | −239.06 | 494.18 | 5.56 | 0.03 |
| Season + f(max 9‐d lag) | 8 | −239.49 | 495.03 | 6.41 | 0.02 |
| Season + f(min 14‐d lag) | 8 | −239.61 | 495.26 | 6.64 | 0.02 |
| Season + f(min avg. 7–14 d lag) | 8 | −239.68 | 495.41 | 6.79 | 0.02 |
| Season + f(mean 7–14 d lag) | 8 | −239.77 | 495.58 | 6.97 | 0.01 |
| Season + f(mean 9‐d lag) | 8 | −239.78 | 495.61 | 6.99 | 0.01 |
| Season + f(max 4‐d lag) | 8 | −239.81 | 495.67 | 7.05 | 0.01 |
| Season + f(mean 14‐d lag) | 8 | −239.86 | 495.76 | 7.15 | 0.01 |
| Season + f(max 10‐d lag) | 8 | −239.98 | 496.02 | 7.40 | 0.01 |
| Season + f(mean 12‐d lag) | 8 | −239.99 | 496.02 | 7.41 | 0.01 |
| Season + f(min 1‐d lag) | 8 | −240.00 | 496.04 | 7.42 | 0.01 |
Only the temperature metric and time lag models with wAICc > 0.01 are shown.
a
Season was a categorical variable defined as the year the samples were collected (November–January).
b
Species was defined as waterfowl or gull.
c
Age was defined as hatch year or after hatch year.
d
Optimal temperature functions were added one at a time to the parsimonious sample covariate model.