How the winner is chosen. Every model is fit by maximum likelihood on the same germination counts, then ranked by AICc — goodness of fit penalized for extra parameters. When two models fit equally well, the simpler one wins. A more complex model has to earn its parameters with a real improvement in fit.
Germination data
Model ranking · by AICc
| Model | k | ΔAICc | weight |
|---|---|---|---|
| Avrami ★ Stretched-exponential kinetics | 3 | 0.0 | 97% |
| Resistant-fraction Persistent non-ageing sub-population | 3 | 7.3 | 3% |
| Ellis–Roberts Symmetric probit decay | 2 | 22.2 | 0% |
| Control-viability Free initial-viability ceiling | 3 | 27.8 | 0% |
k = free parameters. ΔAICc = fit penalty vs best (lower better). weight = probability this is the best model of those tried. ★ = current pick.
VERDICT. Avrami is clearly best for this data (next model ΔAICc 7.3, Akaike weight 97%).
Avrami
A different curve shape entirely — Avrami/stretched-exponential decay, useful when the survival curve isn't a symmetric probit (Niedzielski et al. 2009).
| v0 | 0.952 |
| tau | 1137.1 |
| beta | 0.882 |
| P50 | 1.9yr |
| log-likelihood | -781.2 |
Standardized residuals
Points scattered within ±2 with no pattern = good fit. A systematic arc = the model is missing the curve's shape.
Save & share this model+