FINDING SUSPECTED CAUSES OF MEASUREMENT ERROR IN MULTIVARIATE ENVIRONMENTAL DATA

Environmental data are usually multivariate, with the variables confor ming to some correlation structure. Occasionally, measurements which d o not conform in structure or magnitude may occur in one or more varia bles. It is important (1) to characterize these discordancies in terms of the disturbed variables and the direction and magnitude of the ano malous error and (2) to associate each discordant observation with a s pecific cause of measurement error in order to prevent further mismeas urement. We describe a procedure for identifying suspected causes of d iscordant observations in otherwise multinormal data sets. Variables a re assigned to groups, each of which is associated with a specific cau se of measurement error. Discordant observations are identified with t he generalized distance test or the multivariate kurtosis test. Suspec ted causes of measurement error are identified by repeating the tests with one of the groups of variables omitted in each analysis. The proc edures are evaluated with simulated data sets having a correlation str ucture similar to that of a large environmental data set.