FY 1983

Data Intercomparison Theory—Vol. II, Trinity tests for location, spread and pattern differences

Preisendorfer, R.W., and C.D. Mobley

NOAA Tech. Memo ERL PMEL-39, NTIS: PB83-184101, 91 pp (1982)

In this report, the second in a series of five on data intercomparison theory, we examine three basic measures of data-set separation and procedures by which these measures can be assigned statistical significance. The three measures are for the distances between means (SITES), variances (SPRED), and patterns (SHAPE) of space-time geophysical multivariate data sets. The patterns, in turn, are resolved into spatial and temporal patterns. The problem of determining procedures to generate reference distributions, by which statistical significance is decided, is resolved into five parts, depending on the amount of data available for use. We classify availability of data into five categories: adequate, semi-adequate, borderline, semi-inadequate, and inadequate. For each of these settings we develop procedures to generate reference distributions for SITES and SPRED, and determine the power curves for these two statistics under selected procedures. These power curves are compared with those generated by some classical tests for the relative location and spread of multivariate data sets. The proposed statistics SITES and SPRED and some of their distribution-producing procedures appear to be relatively powerful and robust.