Comparative evaluation of two robust methods for estimating frequency factor and activation energy

Abstract

Of the many methods available for estimating the frequency factor and the activation energy based on rate coefficient data at different temperatures, previous studies have shown that the median method (MM) is less affected by the type of experimental error and outliers. Another method which is robust toward error type is the power transformation weighting method (PTWM). The present study evaluates the precision of linear regression (LR), MM, PTWM, and weighted nonlinear regression (WNLR) with correct weights, through simulation tests, when rate coefficients contain different types of normally distributed experimental error. The results, in general, show that WNLR is the most precise method and that error type affects LR significantly. For applications where distribution characteristics of experimental error are unknown, the two robust methods, MM and PTWM, are found to be preferable. The performance of the former method is slightly superior to the latter for applications where the number of points or range of rate coefficient is small. In other situations, precision of PTWM is marginally better than that of MM. The results further indicate that the performance of PTWM can be improved significantly by increasing the number of points in the data set. However, considering simplicity and robustness toward outliers, MM appears attractive for general usage. © 1988 American Chemical Society.