Final engine tests on a new crankcase motor should be fast and economical yet able to discriminate between oils of high quality. Tests should be properly designed in advance to obtain optimum quantity of information, and the expected variability and its sources should be known. However, the experiment can be designed to partition and measure this variability for use in future experimental designs for the same or similar test methods. This paper describes the method of partitioning and measuring variance due to choice of raters that was encountered in treating data obtained from a taxi fleet test.
