Questions and Answers on the current analysis
    For questions on the OLD analysis go here
 
1- What happened with the so called "Model Independent" results that you showed in the old analysis.
    It was pointed out by the statistics committee (specifically Tom Junk and John Conway) that the so called "Model Independent" section is not model independent at all. Quoting John : "It is an average over models, including ones that make very  little sense or at least have very little physics motivation." We agreed to that interpretation and dropped this section out of the analysis.

2- Didn't you mention  you were going to try a frequentist approach, in addition to the Bayesian one you are currently using ?
  I did mention, but we decided we will not do it. The way we thought we could do it is in the middle of the Bayesian-frequentist combat zone, and so whether that's reasonable or not depends on who you ask. Again, quoting John : " One's temptation is to let the systematic thing (say the acceptance) vary while generating the inevitable pseudoexperiments. After four years of discussing this in the statistics committee I think everyone on the committee agrees this is not the correct thing to do, though it is very seductive. "

3- When the top and Higgs get wide, are the decays handled correctly? (Brig Williams)
   Yes. In this analysis we take all the measures to never fall in a region in which the width of the top or Higgs are greater than 15 GeV (see for example question 7 below). We take care of correcting the efficiencies for large widths. As the MC was generated with a narrow width we can always compute the efficiencies for larger widths.  As to whether the corrections done  agree with what would be obtained by directly generating MC with large widths, the anwer is also yes, because with these  widths (small in comparison to the top mass)  the kinematics of the decay of the  top and Higgs are still similar.
          
4- Why is BR(h0->bb) = 0.9 reasonable for "worst BR combination" result? (Daniel Whiteson)
    Because it equals 0.9 over a large region of MSSM parameter space.

5- On the MSSM limit plots, why do the "error bars" sometimes overlap the exclusion region and sometimes not? (Dan Amidei)
     These are not "error bars". These are 1-sigma values on a-priori expected limit assuming the SM hypothesis.
     The overlap  with exclusion region represents how well the data agree with the SM hypothesis.

6-  Why do we exclude more at low tan beta then high tan beta? (Dan Amidei)
    We simply have more sensitivity there,  where the Higgs decays to eiher csbar or WbB.

7 - Why do you stop at mHiggs = 160? (Jaco Konisberg)
     Because the width of the Higgs gets bigger the higher the mass of the Higgs.

8- Do you need MC at multiple MSSM points? (Daniel Whiteson)
    No. Decay kinematics (and thus efficiencies) don't  depend on stuff like tan beta or mu, just on masses  and widths of particles

9-  What happens to the limit plots as you gather more data?
     The exclusion region creeps toward the middle, and the sensitivity band shrinks.

10- In CDF note 7485, table 4, you show  efficiencies for leptons, including CMUP and CMIO. These two, as an example, are anti-correlated since if the CMUP has a lower efficiency the CMIO eff. is expected to increase.  Do you take this type of correlations into account ? (Stephan Lammel)
    First, let's address the given example. The top dilepton analysis requires CMIO's to be non-fiducial to a muon chamber and CMUP's and CMX's to be fiducial to their respective chambers, so it is not the case that a loss in CMUP or CMX would be accompanied by a gain in CMIO.
   Now let's address the general question. For a given selection (Dilep, LJets1, LJets2+, or LTauH), we break its acceptance uncertainty into a number of pieces --- CMUP ID efficiency uncertainty, CEM ID efficiency uncertainty, etc. These pieces are treated as 100% correlated between two selections if they appear in both selections. But we don't break it down any further than that, e.g. we treat the ID efficiency uncertainties for different lepton species as uncorrelated. Now that's not exactly the right thing to do --- for example, the CMUP ID efficiency uncertainty has some correlation with the CMX ID efficiency uncertainty since they incorporate the same d0 cut. But here's the thing: the effect of correlated systematic uncertainties on the number of expected events gets completely washed out by the (uncorrelated) statistical uncertainties. If we re-did the analysis treating all systematic uncertainties as completely uncorrelated, you'd barely notice the difference! So the fact that we don't break apart systematics into their absolutely fundamental pieces means we're ignoring a tiny correction to an already tiny correction.
11- The Worst Case Combination scenario plot with the sensitivity band (see it here) shows two "non-intuitive" features. A) The first one is that the sensitivity band for high charged higgs masses lies outside the exclusion band (look for example mH=120 GeV). B) The second one is that the sensitivity band at mH+=100 GeV is narrower than that at different Higgs masses. Why is that ? (the authors)
A) Let's analize the first feature. If we look at the distribution of limits for mH=120 ( shown here) there are 9 entries that populate the last bin, which is the one we quote as the worst limit for that mass. That means there are 9 BR combinations that yield the same limit in BR(t->Hb).One of them (picked randomly) is
  • BR(H+->cs)=0.05
  • BR(H+->t*b)=0.7
  • BR(H+->Wh0)=0.0
  • BR(H+->tn)=0.25
    • The number of events in each channel for this particular combination is given in the figure below
    MH100
    The product of poisson distributions (which is the posterior prob.) for this particular number of expected events given the number of events observed (13,8,49,2) and SM expected (11,10,54,2) result in the likelihoods shown in the figure below:
    MH100
    It is clear that data favors the SM hypothesis (i.e. data favors BR(t->Hb)=0), and thus sets a stricter limit than SM itself.

    B) The point at mH=100 GeV is sort of a singular point. The worst limit is given by combinations very similar to what's below :
  • BR(H+->cs)=0.00
  • BR(H+->t*b)=0.9
  • BR(H+->Wh0)=0.0
  • BR(H+->tn)=0.1
    • The number of events in each channel is pretty much flat, and the posterior probability decays only very slowly with BR(t->Hb),regardless of what observed number of events it is compared to in the Poissons. Thus, the 95% C.L. obtained is typically 0.95% of the integration region which ranges from 0 to 0.9. This yields 0.95*0.9= 0.855 which is the limit obtained. The band obtained is narrow because it is insentive to the number of events used in the product of Poissons that make the posterior.


    Go back to main Higgs' Analysis page.