Correlation coefficients assume that the data correlated all arise from the same underlying distribution. This is not true for Fs, since low resolution Fs tend to be larger than high resolution ones. CCs are probably valid in resolution shells. Over all resolution ranges you just see a correlation due the average F effect. This is why in my program Pointless I use CC on E^2, which removes (most of) this effect The correlation coefficient is just the slope of the LSQ line through a scatter plot of (say) F1 vs. F2, so you can see the large resolution Fs will be at the top right of the scatter plot even if they're not the same, small low reso ones at bottom left Phil On 6 Aug 2010, at 20:22, Nathaniel Echols wrote:

On Fri, Aug 6, 2010 at 8:05 AM, James Fraser wrote: What does the correlation coefficient that is reported at the end of phenix.reflection_statistics represent? (I would assume the CC of F1 vs F2 for all reflections with an observation in datasets 1 and 2).

Correct.

Also in the below comparison, how would the overall correlation be 0.901 when all the bins have correlations between 0.330-0.858?

This appears to be a property of correlation coefficients - I'm still trying to figure out the precise mathematical explanation, but it is expected.

http://mathworld.wolfram.com/CorrelationCoefficient.html

-Nat

_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb