The increasing number of model types that are used to predict tree biomass from diameter, height and
wood density has brought questioning about the biological relevance of complex allometries (i.e.
non-power models). Statistical issues such as collinearity among predictors and unreliable coefficient
estimates have also been associated with complex allometric models. Using a data set of 225 trees from
central Africa, we assessed the relevance of simple allometry (i.e. power model) versus complex allometry
to predict tree biomass. A complex allometric model of biomass was developed based on a model
of resource partition between dbh and height growths. Although being a good model for biomass prediction,
the power model was outperformed by the complex allometric model. A careful examination
showed that the power model could be segmented into two pieces of power models. Using tree diameter
and height as separated predictors improved the biomass prediction, irrespective of the collinearity
between these two predictors. A critical value of 25% for the PRSE statistic used to assess the reliability
of coefficient estimates corresponded to a significance level of 105–104 and was thus unreasonably low.
We conclude that growth theories should be developed to explain allometric models, but that the
arbitration between these models should ultimately rely on observed data, not on theories.