Some thoughts on chemical mass balance models

It is argued that multivariate chemical mass balance models are, in general, incapable of determining the number and composition of sources of environmental contamination entirely from the data alone; additional identifying information must be obtained from sources external to the data. Determination of the number and composition of sources is possible, however, if a `libraryʹ of possible source profiles (or partial source profiles) exists. In such a case, a generalized least squares approach to target transformation factor analysis has been shown in the Ph.D. dissertation of H. Yang to have large-sample performance properties that are robust against violations of distributional assumptions. One can alternatively make use of the popular computer program LISREL to fit chemical mass balance models when partial information from a library of source profiles is available. Using the fitted model, it is shown how to obtain point estimates and credible intervals for source contributions (either in absolute terms, or as proportions of total contributions from all sources).