Estimation of attraction models in marketing typically involves the use of the logcentering transformation.. The resultant estimating equations are then linear in the parameters. The log-centering transformation also appears in the statistical analysis of compositional data (CODA). CODA techniques are applied to data on "shares" in a wide variety of disciplines. This paper uses CODA techniques to rationalize the stochastic specification of attraction models. It further shows that another transformation from CODA, the log-ratio transform, can yield simpler estimating equations. The results are illustrated using an empirical example.