The von Liebig, Cobb-Douglas, and Quadratic functions were evaluated u
sing Generalized Least Squares on fertilizer response data for crops g
rown in Ghana and Mall. This was done to determine which model is most
likely to overestimate fertilizer use. Model fit varied for crops and
soil types. The von Liebig was slightly superior, based on prediction
criteria and number of significant variables. The von Liebig function
al form also produced the most conservative optimum yield estimates an
d input levels, while the Cobb-Douglas was most likely to overestimate
yield and hence input use, but produced the highest net returns for c
otton.