Predict Group-Average Activation Fields
Details
Full-response prediction is stricter than extracting a fitted design effect. Every coefficient functional needed for each requested prediction row must be estimable under the complete fitted model. A within-subject term effect and its partial sum of squares can therefore remain estimable when an unrelated between-subject covariate is aliased with the subject intercepts, even though `predict()` refuses the corresponding full response. Reported term effects use only the estimable contribution of the columns removed by that term's hypothesis. Use [estimability()] to compare the `effect_nonestimable_grid_rows` and `prediction_nonestimable_grid_rows` diagnostics before predicting.
Examples
example_data <- design_mfa_example_data()
example_fit <- design_mfa(example_data$Y, example_data$design, ~ saliency + probe)
predict(example_fit, new_design = data.frame(
saliency = ordered(c(20, 100), levels = c(20, 40, 60, 80, 100)),
probe = factor(c("old", "lure"), levels = c("old", "new", "lure"))
))
#> voxel_1 voxel_2 voxel_3 voxel_4 voxel_5 voxel_6 voxel_7
#> 1 1.324822 0.9917741 0.8935248 0.7113824 0.1898471 -0.6709828 -1.615502
#> 2 -1.675080 -1.2420106 -1.0435223 -0.7611432 -0.1395912 0.8210203 1.865286
#> voxel_8
#> 1 -2.32510
#> 2 2.67483