Statistics of phasing
Most useful indicators of the validity of a heavy-atom site are the real occupancy and the atomic thermal parameter(s).
Reduction in RCullis when new site is added is also a good indicator.
RCullis = <P-weighted lack of closure> / <Isomorphous difference>
But for dmin > » 3.5Å occupancies and thermal parameters are highly correlated and the latter should probably not be refined.
To judge a derivative as a whole, check RCullis and the phasing power.
For anomalous data any RCullis is useful.
Phasing power measures the sharpness of the phase distribution (contribution of derivative to phasing information compared with noise).
Phasing power = <Heavy-atom amplitude> / <P-weighted lack of closure>
Note that all root mean-square (RMS) averages are very sensitive to outliers: underestimates RCullis and phasing power, so straight means are safer.
If data accurately scaled, phase of FP is uncorrelated with phase of FH.
Mean figure of merit measures precision of the "best" phase.
<FOM> » mean of cosine(estimated phase error).