where If) is the ith diagonal element of the weight matrix and the units of Prm, are the same as for the/,. The rms residual must be calculated using only observations of the same units. If y contains observations of different types, then a Pmv value may be calculated for observations of each data type.*

Because prna is normalized according to the sum of observation weights, it is frequently more useful than the loss function as a relative measure of the degree to which the solution fits the observed data. However, this parameter alone is insufficient for detecting the two major causes of a poor fit—unmodeled biases and a high level of noise in the observations. Some insight into the contributions of these two phenomena in specific cases is gained by writing pm in the form where k is the weighted mean of residuals, k=-

2 Wt

and op is the weighted rms deviation of the residuals,

0 0

Post a comment