I don't think this is a meaningful question. For Mn(I/sd), we take all measurements of each reflection h to get its average Ih, and an estimate of the SD of this average sd(Ih) (from the adjusted input sigmas), hence the ratio Ih/sd(Ih). Then we average this ratio over all reflections in a resolution bin (or overall) to get Mn(I/sd). This ratio clearly has a distribution which is related to the Wilson distribution of intensities (if all sd(Ih) were the same, it would be the Wilson distribution), ie we have strong reflections (large (I/sd)) and small ones. This SD(I/sd) is not particularly meaningful: it is not an error estimate.
I'm not sure what you mean by "error" in this context. Mn(I/sd) doesn't have an error, it is a characteristic of a particular dataset.
For what it's worth, here is a tabulation against resolution of Mn(I/sd) and SD(I/sd) for a "good-enough-to-be-considered-typical" dataset (created by a program hack)
Phil
dmax Mn(I/sd) SD(I/sd)
6.42 33.26 16.42
4.71 37.85 17.15
3.90 43.29 15.85
3.40 38.46 16.23
3.05 32.53 16.24
2.79 26.42 14.66
2.59 22.43 13.83
2.43 19.89 12.44
2.29 16.88 11.06
2.17 15.49 10.32
2.07 13.59 9.57
1.99 11.29 8.36
1.91 9.08 7.11
1.84 6.63 5.61
1.78 5.06 4.32
Overall 19.17 16.15
On 22 Nov 2010, at 18:33, Bryan Lepore wrote:
> [ scala 3.3.16 ]
>
> in scala's "final table", there's "Mean((I)/sd(I))". i could be wrong,
> but the error of this measurement seems to me to exist, considering
> the uncertainty of sigma = 1 / sqrt( 2 (N-1) ) ... but its not clear
> where the logfile has the values of I or sigma and N that correspond
> to Mean((I)/sd(I)) so i can calculate it myself.
>
> or, am i overlooking a table of perhaps percent data vs. I/sigma in
> scala, or something else...
>
> -Bryan
|