AN ALTERNATIVE VIEW OF THE SPLICING DATA

Data on the time taken to splice telephone cables are re-examined. A p revious analysis using robust regression methods and with the variance a function of x is contrasted with a generalized linear model using a power variance function and a power link with x similarly transformed . The new model is shown to have advantages in allowing the natural mo delling of skew residuals. The dependence of robust regression on assu mptions about the variance function is discussed.