As the phenotypic data contained missing observations, any strain with more than 25% of trait data missing was removed from this analysis. This resulted in the elimination of 32 of the 95 BXD strains, giving us an effective sample size of 63 BXD strains. The resultant data set was subjected to column mean imputation in order to fill in the missing trait values, given that the data were missing at random, i.e. not missing over all measures within a battery. Another issue that was encountered during this analysis was that of dimensionality. The dataset used in the factor analysis had more variables/traits (p) than observations/strains (n) (i.e. n < p). This results in incorrect estimation of the covariance matrix and thereby leads to singularity of the estimated covariance matrix. This issue was addressed by using the James–Stein-type shrinkage estimator (Schafer & Strimmer 2005) of the covariance matrix. R packages e1071 (for missing data imputation), corpcor (for covariance shrinkage), nFACTOR and factanal were used for the purposes of factor analysis. Factor loadings were analyzed to obtain factor interpretations. Factor scores were
