The use of random parameter models in forestry has been proposed as one method of incorporating different levels of information into prediction equations. By explicitly considering the variance-covariance structure of observations and considering some model parameters as random rather than fixed, one can incorporate more complex error structures in analysing data.
Competition indices and variance component techniques were applied to 92 Scots pine (Pinus sylvestris L.) -dominated permanent sample plots on drained peatlands in Northern Finland. By quantifying stand, plot, and tree level variation, it was possible to identify the level (stand, plot or tree) at which the explanatory variables contributed to the model. The replication of plots within stands revealed little variation among plots within a single stand but significant variation occurred at stand and tree levels. Positive and negative effects of inter-tree competition are identified by examining simple correlation statistics and the random parameter model.