Principal components analysis for productive and reproductive traits of Holstein cattle

The application of multivariate methods, such as principal components analysis, can be used to verify latent variables and eliminate poorly representative traits in terms of the proportion of total variance explained. Therefore, the aim of this study was to evaluate the possibility of discarding production and reproduction traits based on principal components analysis in Holstein cattle to eliminate redundant traits and difficult to measure. Productive and reproductive records of 5,217 Holstein females (primiparous and multiparous) obtained between 1998 to 2013 were analyzed. Animals were from a farm located in the Descalvado/ SP, Brazil. The analyzed traits were interval between calving and first estrus (CFE: 59.5?29.5 days), interval between calving and last service (CLS: 158.4?86.9 days), calving interval (CI: 448.7?107.5 days), gestation length (GL: 276.1?5.9 days), 305-day cumulative milk yield (Y305: 9,350.1?2,402.3 Kg), peak yield (PY: 39.7?7.3 Kg), milk production per day of calving interval (MYCI: 23.7?5.2 Kg) and lactation length (LL: 302.0?79.9 days). The consistency of the data was done and outliers were removed from the analysis. Due to differences in the units of variables used, the variables were standardized and a correlation matrix of variables was used to obtain eigenvalues. The analysis was performed by the PRINCOMP procedure of SAS program. From the eight principal components obtained, five of them presented eigenvalue lower than 1.0 and explained less than 20% of the total variance. Consequently, the five traits were subject to discard, given these variables were highly correlated with the principal component of smaller eigenvalue and represented practically insignificant variation. In order of less importance to explain total variance of the dataset, CFE, CLS, PY, MYCI and LL were indicated for discard. The discarded variables were redundant because they presented significant linear correlation with the others. The traits CI, Y305 and GL explained approximately 82% of the total variance of the dataset, therefore they are recommended to be used in future analysis.