Table 1.
Model selection table for the joint-likelihood model, which simultaneously modelled devil density at long-term trapping sites and devil detections on long-term spotlight transects. Here we present the four top-performing models and a null model. We selected the best model based on a leave-one-out cross-validation metric, the conditional predictive ordinate (CPO), with ΔCPO showing the difference from the best model. We present the mean coefficient estimate, with 95% credible interval shown in brackets. ‘nl’ denotes a non-linear effect. All models in this table used the gamma distribution to model density and the negative binomial distribution to model the spotlight counts. See Table S9 for the full model selection table
| Model | −2 × Σ(logCPO) |
ΣCPO | Intercept: spotlight sub-model |
Intercept: density sub-model |
Year | Years since DFTD arrival |
Scaling constant |
% button grass |
% wet eucalypt/ rainforest |
% agric | Gaussian random field |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 7570.5 | 0.0 |
2.61
(−3.46, −1.92) |
−1.58
(−2.3, −0.96) |
nl | nl |
0.70
(0.59, 0.81) |
−1.88
(−3.65, −0.19) |
−0.80
(−1.37, −0.24) |
✓ | |
| 2 | 7571.2 | 0.7 | −2.59 (−3.48, −1.86) |
−1.57 (−2.31, −0.91) |
nl | nl | 0.70 (0.59, 0.81) |
−1.91 (−3.71, −0.18) |
−0.82 (−1.43, −0.20) |
−0.05 (−0.67, 0.57) |
✓ |
| 3 | 8036.3 | 465.8 | −2.64 (−3.80, −1.94) |
−1.53 (−2.42, −0.94) |
nl | nl | 0.76 (0.62, 0.89) |
−1.05 (−1.74, −0.37) |
−0.86 (−1.07, −0.65) |
0.27 (0.01, 0.54) |
|
| 4 | 8038.7 | 468.1 | −2.57 (−3.75, −1.86) |
−1.48 (−2.41, −0.88) |
nl | nl | 0.78 (0.64, 0.92) |
−1.28 (−1.94, −0.61) |
−0.93 (−1.14, −0.73) |
||
| Null | 8884.5 | 1314.0 | −1.07 (−1.12, −1.01) |
−0.49 (−0.60, −0.37) |