Data merging statistics for assessing whether to omit measured intensities

Acta Cryst. (2013). D69, 1204-1214 (http://doi.org/nsn); 1215-1222 (http://doi.org/nsp)

Comparison of refinement statistics against real and simulated data (R_free) with data merging statistics (⟨I/σI⟩ and CC_1/2), plotted against resolution. For this example the observed data are better than simulated data out to 1.95 Å resolution, where CC_1/2 = 0.27 and ⟨I/σI⟩ = 0.9.

These papers highlight again the half-dataset correlation coefficient CC_1/2 as a robust statistic for judging whether to include data to a higher resolution or from other crystals. They show that useful information extends beyond the cutoff point suggested by criteria based on R_merge, which is inappropriate for this purpose. Hard trimming of high-resolution data, omitting useful information, is unlikely to improve the final model, so premature data truncation is a mistake.