Model^{a} | Modifications | Selection criteria | |||||
---|---|---|---|---|---|---|---|

AIC^{b} | BIC^{b} | r_{fit} | r_{LOOCV} | SNR | P_{SW} | ||

1 | Full model | −2653 | −1016 | 0.992 | 0.989 | 9.2 | <3 × 10^{−16} |

2 | Remove terms | −2741 | −1827 | 0.994 | 0.988 | 8.5 | <3 × 10^{−16} |

3 | Remove terms | −2772 | −1988 | 0.993 | 0.990 | 8.5 | <3 × 10^{−16} |

4 | Remove terms | −2772 | −2113 | 0.993 | 0.990 | 8.3 | <3 × 10^{−16} |

5 | Remove terms | −2807 | −2426 | 0.993 | 0.990 | 8.1 | <3 × 10^{−16} |

6 | Remove terms | −2806 | −2446 | 0.992 | 0.990 | 8.1 | <3 × 10^{−16} |

7 | Remove outliers | −2981 | −2621 | 0.993 | 0.992 | 8.7 | 2.1 × 10^{−6} |

8 | Remove terms (“Final model”) | −2983 | −2628 | 0.993 | 0.992 | 8.7 | 2.1 × 10^{−6} |

↵

^{a}Eight iterations of variable selection were performed, each resulting in a different model.↵

^{b}The AIC and BIC metrics were computed from models fit according to the maximum likelihood criterion instead of the restricted maximum likelihood criterion because the former is required for correct AIC and BIC estimates (28).