Based on the assumption that physicochemical properties reflect differences in the size and number of self-associated aggregates of open chain mono-hydroxyl alcohols, six alcohols were chosen from a set of 57 alcohols characterised by 12 physical properties. The aim of the sample selection is to capture the variation of the property matrix in a smaller, represen-Ε½ . tative sub-set. Principal components analysis PCA on the property matrix was used in combination with Soft Independent Ε½ . Modelling of Class Analogy SIMCA to select the six alcohols interactively using score plots. Due to missing values in the property matrix, the available PCA algorithm was compared with a new algorithm that handles missing values correctly.
Ε½ . Because of the chosen pre-processing, the available algorithm could be used. The procedure of sample selection was to i Ε½ . select alcohols which spanned the principal components, ii include an alcohol located close to the origin of the principal Ε½ . component co-ordinate system and finally iii select alcohols which had a large residual with respect to the principal component model of the full property matrix. After the initial selection, the left-out alcohols were tested to be members of the principal component model of the first sub-set. Left-out alcohols found not to be members are to be includedrexchanged in the sub-set. The membership testing is repeated until all left-out alcohols are defined to be members of the final sub-set.