For details on b tagging specific to top analyses see TopBTV.

• b tagging scale factors. The parametrizations and recipe for applying them are described here by the b tagging POG. SFs are available in BTagCalibration tool.
  • There are two sets of SFs, one derived from ttbar events and one derived from b-enriched QCD events. The SFs obtained for QCD events is referred to as "mujets", while the SF combination for ttbar and QCD results is denoted as "comb" in csv files.
    • Measurements sensitive to V_tb (or measuring V_tb) and, measurements that use the same sample as the ttbar-based SFs were derived (essentially all measurements using dileptonic ttbar) should use the QCD-based SFs. When in doubt contact the BTV-TOP contacts.
    • Measurements in tau, single-lepton and all-hadronic channels can safely use the SFs derived from ttbar events.
  • Correction for the b-tag discriminator distribution: if you use the b-tag discriminator distribution in your analysis, e.g., as an input to your MVA (machine learning) algorithm, the full shape of the discriminator has to be corrected rather than the efficiency of a give working point. it is recommended to use the “iterativefit” SFs for this purpose.
  • Efficiencies are dependent on the sample under study as they depend on flavor composition and jet kinematics. If you need the absolute value of the b tagging efficiency, which depends on the physics process you are looking at, you should thus take the value from your MC and apply the provided SF from the POG to it.
• Uncertainties: treat the uncertainties on fake rates and b tag efficiencies in the following way
  • Vary SFlight and SFb/c independently.
  • Vary the sample processes (tt and W+jets etc...) simultaneously.
  • [remove this bullet] This gives positive and negative systematic numbers for SFlight and positive and negative systematic numbers for SFb/c.
  • [remove this bullet] Add the positive SFlight and positive SFb/c systematics in quadrature as well as the negative SFlight and negative SFb/c systematics in quadrature, giving an overall positive and overall negative systematic.
  • For the iterativefit method, BTV provides splitted uncertainties as outlined here.
• Correlations: Currently, there is a preliminary recommendation to combine the uncertainties for b-tagging for the full Run 2 for the fixedWP SFs and the iterativefit SFs. Each analysis aiming at using the full Run 2 data should use the appropriate correlation scheme. Please not that in case of the fixedWP SFs this also comes with an additional breakdown of uncertainties for he SFb/c into a part that is correlated across the years and one that is not correlated across the years.
• Public documents containing more information: BTV-15-001 and BTV-16-002