Z-score prediction matrix

This function takes a matrix of predictive accuracies as produced by get.prediction.matrix and returns, for each subject, a zscore indicating how predictive accuracy from a subject's own embedding relates to those from other subjects/

Usage

z.pred.mat(pmat)

Arguments

pmat

An embedding-to-triplet prediction matrix of the kind returned by get.prediction.matrix.

Value

The predictive accuracy of each participant's embedding in predicting their own triplet decisions relative to the accuracy of other embeddings in the matrix, computed as a z-score.

Details

The matrix diagonal must contain accuracies for each participant's own embedding when predicting their held-out triplet judgments. These diagonal values will be z-scored against the distribution formed by all other accuracies in the row, which reflect accuracies of other embeddings predicting that same participant's triplet data.

If embeddings reflect true, reliable individual differences, a participant's own embedding should predict her held-out judgments better than does another random participant from the matrix. In this case the z-scored self-prediction accuracy will be positive. If the z-score is reliably positive across the whole group of participants, this indicates reliable individual differences in representation.

Examples

#Random prediction matrix
pmat <- matrix(runif(25),5,5)

#Diagonal relatively high
diag(pmat) <- 1 - runif(5)/10

#Zscore diagonal relative to other entries
z.pred.mat(pmat)
#> [1] 1.6647980 1.2075877 3.6377133 0.5696648 1.7409402