The function selects and returns the best GAMM out of the six GAMMs computed in
model_gamm. In the case that the GAMM without any correlation
structure performs best, the output tibble contains the information from
the original model_gam output tibble (therefore needed as input).
select_model(gam_tbl, gamm_tbl)
| gam_tbl | Output tibble from the |
|---|---|
| gamm_tbl | Output tibble from the |
select_model returns the same model output tibble as model_gamm
but with only one final GAMM for each filtered IND~pressure pair.
The best error structure is chosen here based on the Akaike`s Information Criterion (AIC). The GAMM with the lowest AIC value is selected, but only if the AIC difference to the GAMMs with a less complex error structure is greater than 2 (or respectively 4 or 6 depending on the level of nested complexity) (Burnham and Anderson, 2002). Otherwise the less complex GAMM is chosen. The following hierarchy of complexity is considered:
no structure < AR1 < AR2 and ARMA1,1 < ARMA2,1 and ARMA1,2
Burnham, K.P., Anderson, D.R. (2002) Model Selection and Multimodel Inference - A Practical Information-Theoretic Approach. Springer, New York.
Other IND~pressure modeling functions:
find_id(),
ind_init(),
model_gamm(),
model_gam(),
plot_diagnostics(),
plot_model(),
scoring(),
test_interaction()
# Using some models of the Baltic Sea demo data test_ids <- c(67:70) gam_tbl <- model_gam_ex[model_gam_ex$id %in% test_ids,] gamm_tbl <- model_gamm(ind_init_ex[test_ids,], filter = gam_tbl$tac) best_gamm <- select_model(gam_tbl, gamm_tbl)