ABSTRACT

Zero values are often present in a compositional data set, and these are problematic for computing ratios. Zeros are often replaced by small positive values, where the strategy for choosing such values depends on the nature of the compositional data. An alternative is to use a different transformation of the data, which accommodates zeros and is “close” in a certain sense to the logratio one, thus approximating the logratio structure. Closeness can be measured, for example, by how far away a transformation is from being subcompositionally coherent, i.e. its level of subcompositional incoherence. The chi-square distance in correspondence analysis is a possible alternative, often with low subcompositional incoherence, since logratio analysis and correspondence analysis have strong theoretical and practical connections.