Factorizing Yields in Buchwald-Hartwig Amination

The data collected in (1) are revisited and rigorously analyzed. We show that the matrix of descriptors used in the analysis of (1) is up to a linear transformation equivalent to a dummy coded matrix that reflects the design of the underlined experiment. Furthermore, it is argued that the reaction yield is better be modeled as a continuous Bernoulli random variable. Finally, a four-way ANOVA model with single replicates following a continuous Bernoulli distribution is fitted to the data under the assumption of the sparsity of the parameters, and estimated parameters are interpreted. Thereby, a novel regularisation technique based on the partial least squares algorithm is applied.