In the original fit of the 2006 data it seems that they have considered the shifts of all 3 data sets as independent.
For the normalization error, the normalization error is supposed to be the same for both 820GeV samples, and is
p.errs[0] (820GeV) = sqrt(pow(5.8,2) + pow(1.5,2) + pow(1.0,2));
And for sample with 920GeV:
p.errs[1] (920GeV) = sqrt(pow(7.4,2) + pow(1.5,2) + pow(1.0,2));
The normalization unc. of both 820 samples is assumed to be correlated.
In the original fit of the 2006 data it seems that they have considered the shifts of all 3 data sets as independent.
For the normalization error, the normalization error is supposed to be the same for both 820GeV samples, and is
p.errs[0] (820GeV) = sqrt(pow(5.8,2) + pow(1.5,2) + pow(1.0,2));
And for sample with 920GeV:
p.errs[1] (920GeV) = sqrt(pow(7.4,2) + pow(1.5,2) + pow(1.0,2));
The normalization unc. of both 820 samples is assumed to be correlated.