Statistical techniques and Doppler satellite positioning
Two problems are addressed: ( a ) the detection of outliers in (Doppler satellite) observations, and ( b ) the testing of coordinates in (Doppler satellite) networks. In both problems, confidence regions of the ‘out of context’ and ‘within context’ varieties are developed, and it is shown that the latter are in general about 1 1/2 times larger than the former (conventional) confidence regions. On the basis of this comparison, it is speculated that good data and results are being erroneously rejected. Also it is demonstrated, through the use of Bonferroni’s inequality, that discarding covariances among residuals and discarding cross-covariances among station coordinates each results in a confidence level being greater than 1 — x, the conventionally chosen level. As a final development, a link is made not only between univariate and multivariate testing for outliers among observations but also between testing in observation space and testing in parameter space. The implications of these developments for Doppler satellite positioning are given.