Sentence examples for predictor pool from inspiring English sources
Because of the lack of robust bootstrap correlation with math, the Stop Signal task was omitted from the initial potential predictor pool.
The DAT concentration in the remaining ROIs was evaluated using a backward elimination general linear model (GLM) regressed on the initial predictor pool: diagnostic group, age, education, putamen DAT, duration of motor symptoms, LED and the interaction of diagnosis with each of the other predictors.
To exclude the possibility that the GCT covariate was redundant and multicollinear with DAT concentration, given that the altropane retention data were already partial volume corrected (using local rather than global volume), we reran all the analyses without including GCT in the GLM's initial predictor pool, and the results were virtually unchanged.
Results suggest that models with limited predictor pools can substitute other predictors to compensate for the missing variables.
The performance for the models built from the different starting predictor pools was analyzed for seven target variables.
Next, the predictor variable pool is increased by adding climate signals (Southern Oscillation Index (SOI), Pacific Decadal Oscillation (PDO)) and Pacific Ocean SSTs from regions that were determined to be teleconnected with UGRB SWE.
Table 2 Multivariate results for random-intercept regression models of short returns Predictor I. Pooled II.
Table 3 Multivariate results for random-intercept regression models of remittances Predictor I. Pooled II.
Small differences were observed between measures based on pooled linear predictors versus pooled probabilities; we expected this finding for the AUROC measures as the estimates are based on ranks of the predictions and not their absolute values.
The consistency of these common and basic predictors facilitates pooling and comparisons of data and interpretation of results from the two countries.
Removing those variables that had p >.10 resulted in a reduction of pool predictors from 48 to 12 for PCS and from 48 to 17 for MCS, thereby reducing the probability for type I errors.
