electron scale factor

[gerbaudo]

Trying to understand the fake electron scale factor. Select events with MeasureFakeRate2::passHFCR_testSs Try to address the following questions:

• do we have enough events to determine a scale factor?
• what type of leptons (real/conv/hf/lf) do we have for each category?
• how can we subtract the true lepton contamination? (just MC? iterative?)

[gerbaudo]

Two plots; more plots here Number of raw events

Mar_27
elec_fakeHF_all_onebin_[num,den]
sample           num      denom
allBkgButHf    37316.0    72731.0
allBkg         37319.0    72861.0
data            4095.0    65598.0
diboson        35724.0    54432.0
heavyflavor        3.0      130.0
higgs           4187.0     6842.0
ttbar           1439.0    15263.0
wjets             44.0     1722.0
zjets            109.0     1314.0

[gerbaudo]

note to self:

LEP_TYPE="conv hf hflf lf real"
for L in ${LEP_TYPE} ;
do
./python/plot_histo.py -n elec_fakeHF_all_l_pt_${L}_num -o trashbin/fake_el_sf_test/ out/fakerate/merged/*${TAG}.root
./python/plot_histo.py -n elec_fakeHF_all_l_pt_${L}_den -o trashbin/fake_el_sf_test/ out/fakerate/merged/*${TAG}.root;
done

[gerbaudo]

Things to check:

• how much does the SF change when derived in the SS region? will probably have to do this in one eta bin, and perhaps one/two pt bins (low stats). will probably also need to just perform a subtraction of the real lepton without any iterative procedure, to keep it simple. For this you need to:
  • [x] re run the old passHFCR with the additional 'mc-truth-labeled' histograms (tag Apr_01)
  • [x] re run the old passHFCR && SS with the additional 'mc-truth-labeled' histograms (tag Apr_02)
  • [x] write a little py script to do the subtraction and compute the ratio
• eta dependence. Is there any eta dependence in the scale factor? this is becoming very unpractical with Matt's histogram-based code. Need to:
  • [x] write out small trees that contain tag&probe lepton, one tree per selection region can be based on the TupleWriter written for the optimization
  • [x] write py code to loop over these trees and fill efficiency histograms
• Determine whether the problem is really with the HF/LF scale factor. This is more complicated. To me, it looks like this could be to conversions in the forward region instead. However, because of the per-sample parametrization we do, it's difficult to pick this up (we quickly run out of stats). Also, intuitively the detector doesn't care whether the lepton is from say ttbar, W+jets or whatever. Again, I think we should move away from the weighted average matrix method. In the meantime, I should probably:
  • [ ] save additional variables for the lepton, whatever is available within SusyNt (d0_sig, iso, ...)
  • [ ] Andy also suggested to look at the b-tag output for the jet that overlap with the electron (that jet is thrown away by the overlap removal). This could help understanding whether the electron could be from b->W+q->q+e+v (could this be enhanced in the forward region?)
  • [ ] go back to the d3pd, and look at more basic variables. For example: there is more material in the fw region. Could these electrons be early-showering jets? look at the em fraction, number of nearby tracks, isolation
  • [ ] you can start by doing this on data only, selecting the handful of suspicious events