Compare the convergence of the model using the canopyDistance predictor as the competition effect:

  • sim1: sigmoid with Lo [0-1] and mid [0.1 - 0.5] parameters
  • sim2: sigmoid with Lo [0-1] and beta [-10 - 10] parameters

Rhat

Divergent transitions

Density plot of each parameter by species id

Data distribution

Size, temperature, precipitation and competition effect

Predictions

Out-of-bag predictions

To compare the model efficiency, we plot the 1:1, and the regression line from lm(pred_growth ~ observed_growth) for both Bayesian and random forest methods as reference.

Mean squared error (MSE) of out-of-bag predictions

Density distribution draws from the posterior distribution (\(n = 200\)) can be compared with the MSE values using the random forest methods (vertical bars). Furthermore, I added the sim_full MSE when running the random forest with all non-correlated variables to get the maximum explicability from data.

Rsquared

Rsquared values draw from the posterior distribution is calculated using the Gelman et al. 2018 definition. Like MSE, the \(R^{2}\) can be compared with the ones from the random forest using the same variables along with the full model (sim_full). However, the \(R^{2}\) of the random forest is calculated using the classical equation.

Intraclass Correlation Coefficient (ICC)

Population summary

Sampling time