| Abstract: | Four topics, selected from the range of statistical subjects with current relevance for agricultural research are considered. The first is concerned with the need and usefulness of efficient incomplete block designs for any number of treatments, either equireplicate or with different number of replicates. The second one is a presentation of linear prediction procedures for phenomena with spatially random behaviour and its generalization towards kriging. Next, a reparametrization of a general four–parameter family of growth curves is recommended, leading to numerically stable parameter estimation for RICHARDS' functions and additional special cases. Finally, the modelling of factorials with continuous response and of log–linear relationships for frequencies is proposed to be handled on the same footing, with preference for an alternative criteria of model deviance over test critera or Cp–like measures. |
