ABSTRACT
A graph of the residuals versus fitted values gives no evidence that the error of log cycles to failure varies with the mean response: thus the data seem reasonably consistent with the central assumption of accelerated life testing. Cycles to failure vary over a very wide range and the amount of random variation is likely to increase with the mean cycles to failure. There are a number of reasons why use of log cycles to failure is likely to be the most effective way of analysing these data. The general form of the systematic variation can be studied very easily and directly from appropriate mean values collected in two-way and one-way tables, as for other forms of balanced factorial experiment. While the final summary of conclusions is likely to be primarily in terms of a fitted regression equation critical inspection of two-way tables is all the same an important intermediate step in the analysis.
