Home » Groups » STAR Protected

Update 09.13.2018 -- Run 9 pp: Unfolding Errors

Updated on Wed, 2019-02-13 15:55. Originally created by dmawxc on 2018-09-13 11:31.

A question that's been dogging us for a while is how to disentangle the errors due to the unfolding from the propagated statistical errors.  Unsurprisingly, RooUnfold has methods built in to keep track of things; surprisingly, I didn't realize this until yesterday.  Here're the (absolute) errors due to the unfolding for the 'eTtrg = 11 - 15' GeV pi0 spectrum (R = 0.2, charged jets):

For reference, this is for the Bayesian algorithm with 'k = 4'.  To get an idea of the size of the errors, the table below compares the percent-error from the unfolding to the total percent-error of the unfolded spectrum.  The total percent-error, which is all statistical here, on the measured distribution is also included.

bin no. %-error [unfolding] %-error [total] %-error [total, measured]
9 1.07 3.61 1.24
10 1.59 4.07 1.68
11 2.02 5.85 2.06
12 2.68 6.08 2.86
13 2.86 7.00 2.83
14 3.29 6.51 3.05
15 4.13 7.54 3.86
16 5.70 8.94 4.99
17 7.73 10.53 6.80
18 9.97 13.83 9.67
19 2.03 22.29 21.82
20 9.28 14.03 n/a
21 25.83 37.17 n/a

However, if you look at slide 14 of my September 11th presentation in JetCorr, the errors on the unfolded distribution look miniscule:

https://drupal.star.bnl.gov/STAR/system/files/JetCorrUpdate_Anderson.Sep18.v2.pdf

They look smaller than the measured errors, in fact.  This is just a quirk of the way I plotted it; if you plot as a binned distribution, it looks completely reasonable.

The error bars are reasonably sized, but when I draw it as a band, the large bin sizes make them look much smaller than they actually are.
Update [09.13.2018]: not that it really matters here, but this test was actually done on a pi0-triggered spectrum.  The text and plot label stated that it was a gamma-rich-triggered spectrum; the text has been changed, but please ignore the plot label.

Groups: