Empirical assessment of the relative performance of orthogonal contrast analysis for optimization

Optimization in decision making is essential for efficient allocation of rare resources. Finding the optimal values for a given process has been a key issue in statistical science. Analysis of variance is one of the popular statistical methods used to test significance of the differences between means of levels of a factor or combinations of levels of multiple factors while contrast analysis aims at identifying the fine structure of those differences. Specifically, contrast analysis aims to find the optimal value of a response variable and the consequent value of the predictor. Orthogonal contrasts analysis (OCA) is the commonly used contrast method. Performance of OCA could however depend on the number of levels of the factors tested. Using simulation, we propose to assess the effect of the number of levels of a factor, the degree of non-normality (normality, moderate, severe, and very severe) and heteroscedasticity (homoscedasticity, moderate, severe, and very severe) combined with the three equations of type 1 error correction on the precision of the OCA. Findings will be used to identify the best practices for optimal baobab leaves’ production in smallholder farming systems in Benin. The identified optimal practice will be recommended for adoption by extensionists as part of dissemination to communities through farmers’ organizations.