A critique of some regression adjustments used in allometric size “correction” in numerical taxonomy



Residuals from an allometric regression line fitted in log–log space between linear measurements and a size variable are frequently used in numerical taxonomy as “size-adjusted” variables. Two forms that these residuals can take are: (1) log observed – log expected and (2) following conversion back to original scale, (observed – expected)/expected. Clustering and ordination procedures are used to demonstrate that summaries of morphometric affinities can be influenced by choice of residual. The discrepancy is predicted to be greater when some objects of study have predominantly positive or negative residuals. This situation may be more likely when many variables are functionally related. An argument is made for use of log residuals. Among the many other size “corrections” employed in multivariate morphometrics are those of Corruccini (1972,1975a, 1978). These are examined and are shown to give undesirable results.