Table 1.
Prediction squared error (PSE) or classification squared error (CSE) in out-of-sample prediction using r with different proportions of variance explained
Proportion of variance explained by PCs used to compute  (%) |
RR parameter | ||||||
|---|---|---|---|---|---|---|---|
| 10 | 50 | 70 | 90 | MAX | CV |  |
|
| Continuous outcomes (mean PSE) | 1.24 | 1.23 | 1.23 | 1.27 | 3.20 | 1.24 | 1.23 |
| (sd) | (0.06) | (0.05) | (0.05) | (0.06) | (0.87) | (0.05) | (0.05) |
| Binary outcomes (mean CE) | 0.46 | 0.47 | 0.47 | 0.47 | 0.47 | 0.47 | 0.46 |
| (sd) | (0.03) | (0.03) | (0.03) | (0.04) | (0.03) | (0.03) | (0.03) |
MAX, r is the maximum number of PCs where the corresponding eigenvalues are non-zero; CV, r chosen using cross-validated PRESS statistic; 
: r chosen based on degrees of freedom (see main text); SD, standard deviation.