Our main approach consisted in fitting multivariate biometric models using a Cholesky parameterization. This permitted us to determine whether adding candidate endophenotypes to our model would account for significant unique variance in EXT. Because our approach to decomposing time-frequency surfaces differs from that used in previous studies relating TF energy to externalizing psychopathology, yielding relevant dimensions not necessarily readily evident to the naked eye (Gilmore et al., in press), it was not immediately apparent which features would be most relevant. We therefore conducted principal component analysis of the 15 TF-PCA measures (5 components, 3 electrodes) and EXT to identify TF-PCA components that are related to (load with) EXT (c.f. Patrick et al., 2006). We used the components that loaded with EXT in subsequent biometric models. PCA and all data manipulations were conducted using R (version 2.8.1; R Foundation for Statistical Computing).
