Panel data validation using cross-sectional methods.

Abstract

An examination of the suitability of an Irish Sitka spruce research panel data set for growth modelling purposes was undertaken. The panel data set arose from several repeated measurements in replicated field experiments. When being considered as data for yield modelling several difficulties arise. Simple histograms comparing sampled plots and the underlying forest estate demonstrated a sampling imbalance - whereby site index classes for sampled plots misrepresented the population. The spatial proximity of established plots also meant there was a lack of randomisation at the plot level, which eroded statistical independence between plots and increased plot cross-correlations. However, the availability of independent, non-research volume data permitted the construction of stand-level volume equations for both research and non-research stands. Observed differences in volume equation residuals for research thinned and unthinned stands were then explored. Thinning effects, volume equation inadequacy, or other sampling biascs were considered as potential candidates to explain residual differences. It was found that the differences were consistent with a form of sampling bias when measuring volume sample trees. These validation techniques have led to a better understanding of the research data set.