Modeling wood quality using random regression splines
During the last five years there has been a surge of wood quality data describing variation from pith to bark. This data has been generated by low-cost repeated assessment of individuals or by SilviScan-type analyses of increment cores. Traditionally, breeders have used multiple univariate analyses to study the evolution of genetic control and estimate age-age correlations. In this paper I random regression splines to model longitudinal data. After introducing theoretical considerations, I apply the method to two SilviScan data sets from open-pollinated progeny trials, using microfibril angle (°) for 188 trees from an 11 year old Eucalyptus globulus trial in Australia and basic density for 425 trees from an 28 year old Pinus radiata trial in New Zealand. The random regressions approach permits modeling the mean wood quality curve and individual curves at the family and tree level. These curves show significant variation from the mean, which can be exploited for selection purposes, particularly when quality thresholds are important. In addition, their use produces a smoothed additive genetic covariance matrix.