We showed that when a limited number of predictors were included in the BRT model, the standalone BRT predictions were significantly improved by robust geostatistical modelling of the residuals.
No predictors were included to explain the variability in the slope.
Predictors were included in each model analysis, and a backward conditional method was used for variable selection using logistic regression (categorical outcomes) or linear regression (continuous outcomes).
The analyses are "complete cases" analyses, that is, only participants with non-missing data on both the outcome and the 6 predictors were included.
Therefore, in the final models, the slopes associated with these level-1 variables in the U.S. and Japan were allowed to vary randomly across schools, but no level-2 predictors were included in the models to predict that variability.
When domain-wide predictors were included in a final model, independent predictors of poor sleep quality in the BD sample included rapid cycling, high neuroticism, and undesirable events in the past 6 months, and those in the HC sample included social stress.
Six predictors were included in our model which were related to land cover (including two NDVI-related predictors) and one related to soil permeability (Table 1).
All forms of two-factor interactions, as well as three-factor interactions between the three predictors were included in the model.
We used forced-entry methods to include the allocated group and factors that may not have been well balanced between arms at baseline, while other potential predictors were included based on a P-value<0.20 in univariate analysis and were selected using a backward procedure.
In the active guessing and passive observation runs two predictors were included in the general linear model: (a) the period in which the movie was shown until button press and (b) from button-press until the participant had given the answer.
All predictors were included in the analyses.
