SIMPLE ROBUST-TESTS FOR SCALE DIFFERENCES IN PAIRED DATA

The classical statistical test for assessing the difference between th e two scale parameters in paired data due to Pitman (1939) and Morgan (1939) is not robust. This paper explores robust alternatives to Pitma n's test, using the framework of the one-sample t-test. The asymptotic behaviour of the tests under the null hypothesis is examined. In gene ral, it is easy to construct tests that are asymptotically distributio n-free, provided the data come from a symmetric bivariate distribution . Versions of these tests that do not require symmetry are derived. Mo nte Carlo simulation experiments are done to assess small sample behav iour and power characteristics. The tests are applied to a small data set comparing cancerous and noncancerous lungs.