. 2022 Mar 9;24(3):386. doi: 10.3390/e24030386

Table A16.

Triple of simulated prediction interval lengths averaged over the five different covariance structures for the linear model and $r = 20 %$ missing rates. The triple covers the sample sizes $(n_{1}; n_{2}; n_{3}) = (100, 500, 1000)$ using a significance level of $α = 0.05$ .

Prediction Interval Length for the Linear Model with r = 20% Missings
Imputation Method	$P I_{n, e m p Q}$	$P I_{n, R e s V a r}$	$P I_{n, w e i g h t e d}$
missForest	$(12.214; 11.953; 11.840)$	$(12.701; 12.053; 11.886)$	$(11.997; 11.149; 10.896)$
mice_pmm	$(12.457; 12.315; 12.237)$	$(12.969; 12.411; 12.276)$	$(12.263; 11.503; 11.281)$
mice_norm	$(12.420; 12.253; 12.160)$	$(12.930; 12.344; 12.201)$	$(12.226; 11.440; 11.210)$
mice_rf	$(12.519; 12.347; 12.267)$	$(13.017; 12.439; 12.297)$	$(12.310; 11.529; 11.301)$
gbm	$(12.283; 12.001; 11.857)$	$(12.785; 12.099; 11.903)$	$(12.079; 11.194; 10.913)$
xgboost	$(12.269; 12.003; 11.875)$	$(12.769; 12.102; 11.918)$	$(12.063; 11.196; 10.927)$