-- AmnonHarel - 12-Sep-2010

How should we interpret the "disappearance" of the CLs limits at high lambda?

CLs basics

The "modified frequentist" CLs approach is to exclude regions of phase space where $CLs=\frac{CL_{s+b}}{CL_b} < 1-\alpha$, where $\alpha$ is the desired confidence level. Here, $\alpha=0.95$.

Fine print

None of these really matters at the end, but for completeness:
  • When the number of pseudo datasets (PDSs) that underlie either $CL_{s+b}$ or $CL_b$ at the limit is below 10, we consider the determination of the limit unreliable and do not quote it.
    • For the final results, we generate enough PDFs at the crucial points in phase space to prevent this requirement from effecting the results, except where it clearly removes statistical noise (see $\lambda=5$TeV example below)
  • Since the test statistic is the log likelihood ratio, with or without systematics, when we define an excluded region of the LLR, it will normally be simply connected and include $-\infty$ (the sign convention is such that this is the value most unlike the new physics scenario). We considered the possibility that when the LLR does not include systematics, this may fail to hold, in which case we would quote only the uppermost excluded LLR in the lowest exclusion region. Currently, we have no reason to believe this abnormal situation arises in our measurement.

Final results

What changed

  • Found a bug that greatly reduced the shift uncertainty in all results, and another that reduced the absolute JES in the results sent to the statistic board during CWR. The three leading uncertainties were unaffected by these bugs, so the results barely change. It is only due to the finicky nature of CLs limits in the regime we're at, that these tiny effects must be resolved.
  • Better understanding of the low LLR tails, described in the next section
  • Now using the CMS LPC batch system - generating obscene amounts of pseudo-datasets has never been easier smile

What drives the low LLR tails?

The earlier results:
  • the nuisance parameters in the tails are as one would expect from their correlations with LLR in the bulk. They are only mildly effected.
  • the plot of the ensemble broken down by the last non-empty inner bin:
    more_ex1.png
both hinted that statistical effects drive the low tails.

To check this further, we took a key ensemble and filtered out PDSs where any of the nuisance parameters is more than 2 sigma away from its nominal value, and checked how this effects the LLR tails.
plot_syst_tail.png
To see the key numbers more clearly, here are the relevant table entries (tables explained below):

  Lambda N_{PDS} At the CLs point Frequentist limiting Data
  N_{S+B} CLb CLsb LLR value LLR value LLR value N_{S+B} CLs MC stat. err. on CLs
Full systematics 4.00 151949 104 0.0007 0.0137 -3.38 -1.67 -3.57 40 0.0415 0.0071
Truncated systematics 4.00 192126 86 0.0004 0.0089 -3.50 -1.69 -3.57 55 0.0451 0.0063
The effect on CLs is negligible.

We conclude that the low LLR tails are driven entirely by the statistics, and our ability to study them is not limited by the accuracy of the systematic uncertainties.

Plots

The dashed red line in the right hand plots is the CLs=5% line.

Lambda LLR distributions CLs plot
2TeV cpsdn20001I2_88.png v7ex1.png
3.6TeV cpsdn36001I2_88.png v7ex8.png
4.0TeV cpsdn40001I2_88.png v7ex6.png
4.05TeV cpsdn40501I2_88.png v7ex5.png
4.1TeV cpsdn41001I2_88.png v7ex7.png
4.15TeV cpsdn41501I2_88.png v7ex9.png
5.0TeV cpsdn50001I2_88.png v7ex2.png

  • The two peak structure visible at 5.0TeV is due to having one very high mass event, which can be either inner or outer. See also the "old" plots of 4.05TeV below, with a breakdown by the last non-empty inner bin.

The table

"N_{S+B}" is the number of events in the new physics (QCD+contact interaction) ensembles whose LLR is below the relevant LLR value (CLs limiting value or observed data value).

Lambda N_{PDS} At the CLs point Frequentist limiting Data
N_{S+B} CLb CLsb LLR value LLR value LLR value N_{S+B} CLs MC stat. err. on CLs
1.80 4000 199 0.0497 0.9943 -1.09 -1.06 -27.75 1 0.0009 0.0009
1.85 4000 198 0.0494 0.9888 -1.53 -1.46 -25.50 1 0.0010 0.0010
1.90 4000 197 0.0493 0.9854 -2.02 -1.93 -24.14 1 0.0011 0.0011
1.95 4000 196 0.0490 0.9791 -2.18 -2.08 -23.32 0 0.0005 0.0005
2.00 4000 194 0.0486 0.9720 -2.38 -2.24 -21.23 0 0.0000 0.0000
2.20 4000 181 0.0453 0.9069 -2.79 -2.49 -15.19 1 0.0019 0.0019
2.40 4000 180 0.0449 0.8975 -2.05 -1.77 -13.98 0 0.0000 0.0000
2.60 4000 155 0.0387 0.7745 -2.55 -1.99 -10.90 0 0.0000 0.0000
2.80 4000 125 0.0312 0.6244 -2.80 -2.02 -8.84 0 0.0000 0.0000
3.00 4000 88 0.0220 0.4395 -3.01 -2.09 -7.02 0 0.0013 0.0013
3.20 4000 44 0.0110 0.2190 -3.64 -2.09 -5.99 1 0.0101 0.0101
3.40 30000 179 0.0060 0.1191 -3.41 -1.79 -4.86 13 0.0276 0.0079
3.60 244084 455 0.0019 0.0373 -3.50 -1.64 -4.00 120 0.0373 0.0035
3.80 92098 187 0.0020 0.0406 -3.01 -1.43 -3.66 23 0.0321 0.0068
4.00 402210 155 0.0004 0.0077 -3.53 -1.69 -3.57 125 0.0481 0.0044
4.05 323618 --- --- --- --- -1.65 -3.41 133 0.0627 0.0056
4.10 139028 45 0.0003 0.0065 -3.27 -1.62 -3.27 43 0.0491 0.0077
4.15 152000 19 0.0001 0.0025 -3.34 -1.63 -3.18 49 0.0530 0.0079
4.20 111028 3 0.0000 0.0006 -3.32 -1.55 -3.03 42 0.0690 0.0113
5.00 25000 --- --- --- --- -1.08 -1.49 48 0.2426 0.0418

Conclusions

Going over the crucial lambda values one by one:
  • Though the separation at lambda=3.6TeV is low enough that it is almost power-constrained away, this lambda value is clearly excluded.
  • 3.8TeV is excluded
  • A CLs crucial value exists at 4.0TeV. It is probably above the data so that 4.0 is excluded. Generating more PDSs...
  • 4.05TeV?
  • A CLs crucial value exists at 4.1TeV. excluded?
  • 4.15TeV?
  • 4.2TeV?
  • If a crucial CLs value exists at 5TeV, the data value is well above it and 5TeV can not be excluded.

Updated results during CWR

Generated an obscene amount of additional PDSs...

Tools

  • Added statistical uncertainties on the CLs values.
    • Since the # of PDS ($N$) is predetermined, the uncertainties on CLb and CLsb are (independent) binomial efficiency problems. I use the usual approximation of putting in the observed efficiency ($p=A/N$, where $A$ is the number of PDSs below the LLR value). This gives an uncertainty on the relevant fraction ($f$, either CLb or CLsb) of $\sigma_f = f \sqrt{ \frac{1}{A}-\frac{1}{N}}$.
    • Then the two numbers are combined with standard error propagation, which implies the Gaussian approximation.
  • Added more diagnostic columns to the tables: N_{S+B} (explained below), CLs at data with its uncertainty due to MC statistics.
  • Plots now available for all relevant lambda values

Plots

The dashed red line in the right hand plots is the CLs=5% line.

Lambda LLR distributions CLs plot
2TeV cpsdn20001I2_88_v4.png n8ex1.png
3.6TeV cpsdn36001I2_88_v4.png n8ex8.png
4.0TeV cpsdn40001I2_88_v4.png n8ex6.png
4.05TeV cpsdn40501I2_88_v4.png n8ex5.png
4.1TeV cpsdn41001I2_88_v4.png n8ex7.png
4.15TeV cpsdn41501I2_88_v4.png n8ex9.png
5.0TeV cpsdn50001I2_88_v4.png n8ex2.png
  • The two peak structure visible at 5.0TeV is due to having one very high mass event, which can be either inner or outer. See also the "old" plots of 4.05TeV below, with a breakdown by the last non-empty inner bin.

The table

"N_{S+B}" is the number of events in the new physics (QCD+contact interaction) ensembles whose LLR is below the relevant LLR value (CLs limiting value or observed data value).

Lambda N_{PDS} At the CLs point Frequentist limiting Data
N_{S+B} CLb CLsb LLR value LLR value LLR value N_{S+B} CLs MC stat. err. on CLs
1.80 4000 199 0.0498 0.9963 -1.01 -0.99 -27.75 0 0.0000 0.0000
1.85 4000 199 0.0497 0.9939 -0.82 -0.78 -25.50 0 0.0000 0.0000
1.90 4000 197 0.0494 0.9875 -1.49 -1.41 -24.14 0 0.0005 0.0005
1.95 4000 198 0.0495 0.9906 -0.78 -0.73 -23.32 1 0.0015 0.0015
2.00 4000 196 0.0489 0.9787 -2.04 -1.92 -21.23 0 0.0002 0.0002
2.20 4000 185 0.0462 0.9238 -2.24 -2.03 -15.19 0 0.0000 0.0000
2.40 4000 177 0.0443 0.8856 -2.17 -1.83 -13.98 0 0.0000 0.0000
2.60 4000 184 0.0460 0.9194 -0.29 -0.12 -10.90 0 0.0000 0.0000
2.80 4000 166 0.0414 0.8276 -0.94 -0.71 -8.84 0 0.0000 0.0000
3.00 4000 84 0.0209 0.4188 -3.10 -1.95 -7.02 2 0.0211 0.0162
3.20 4000 63 0.0157 0.3132 -2.99 -1.81 -5.99 2 0.0181 0.0148
3.40 4000 19 0.0047 0.0933 -3.46 -1.67 -4.86 1 0.0265 0.0268
3.60 21000 121 0.0058 0.1153 -2.74 -1.46 -4.00 10 0.0389 0.0123
3.80 12000 28 0.0023 0.0466 -2.90 -1.38 -3.66 2 0.0261 0.0187
4.00 16000 32 0.0020 0.0395 -3.02 -1.47 -3.57 1 0.0057 0.0057
4.05 181000 346 0.0019 0.0382 -2.89 -1.47 -3.41 44 0.0403 0.0063
4.10 146000 178 0.0012 0.0244 -2.91 -1.46 -3.27 35 0.0434 0.0076
4.15 35000 --- --- --- --- -1.46 -3.18 15 0.0793 0.0216
4.20 4000 --- --- --- --- -1.61 -3.03 4 0.1469 0.0859
5.00 4000 --- --- --- --- -1.07 -1.49 11 0.4184 0.1493

Conclusions

Why did we see fake rises?

  • the main problem is that we evaluated the significance using the number of PDS in the low-end rise region. The right way to think of it is by analogy to a bump hunt - is there a signficant excess over background? Just using the number of PDS is looking at S+B. In particular, a high B is a reason to distrust an excess, not to trust it frown
    • it also hurt that the S+N numbers weren't readily available. They are in the tables now (they mostyle help answer other questions).
  • also a mild "look elsewhere" effect

Which CLs limits should we use?

It is not safe to use CLs limits which are based on small tail probabilities. How small?
  • From these results, anything below ~0.001 is suspect
  • proved irrelevant - see above! Our description of systematic uncertainties wasn't meant to cover such extreme cases. It's probably good enough till at least 3 sigma --> ~0.0015. It's certainly not good enough at 4 sigma (truncated there) --> ~0.00003. Also the choice of the prior for the JES wasn't checked below ~0.001 (see AmnonHarelStatusReportForStatisticsBoard#New_prior_shapes_for_nuisance_pa)

Going over the crucial lambda values one by one:

  • Though the separation at lambda=3.6TeV is low enough that it is almost power-constrained away, it is certain that a CLs limit exists and the data lies below it, so this value is excluded.
  • 4.0TeV is excluded
  • A CLs limit exists at 4.05TeV. It is probably above the data, so that 4.05 is excluded. I'd guestimate the significance of the previous statement, in terms of MC statistics, at around 2 sigma.
  • A CLs limit exists at 4.1TeV. There is a hint that it is above the data, so that 4.1 is excluded. I'd guestimate the significance of the previous statement, in terms of MC statistics, at around 0.5 sigma.
  • There is a hint that no CLs limit exists at 4.15TeV. If a crucial CLs value exists, the data value can easily (~50%?) lie above it.

So the best choice is to stick with the limits presented in the CWR, and add one more item to the discussion of the difficulties with CLs:

"CLs sometimes requires estimating extreme tails which stresses the validity of the systematic variations and is difficult using brute force ensemble testing."

the above may be true, but we now (after CWR) realize it is irrelevant to this measurement

Need to decide on the right presentation though...

Improved plots with ensembles as of a bit before CWR (early 21st of September)

The plots on this section use systematics which are slightly lower than the final ones. There are not qualitative differences, in particular, the sensitivity runs out at 4TeV either way, and quantitative differences are likely to be tiny.

The simple cases

To understand how the CLs limits go away, we'll look at plots of CLs as a function of LLR (for a given new physics model, here contact interactions).

Easy exclusion

LLR distributions CLs plot
cpsdn20001I2_88_v2.png ex1.png

No exclusion

With systematics (as usual) Only statistical variations
LLR distributions CLs plot LLR distributions CLs plot
cpsdn50001I2_88_v2.png ex2.png cpsdn50002I2_88.png cls_wo_syst_5TeV.png
The leftmost "0" is real, but due to a single point from the SM ensemble. The 0,0 points (quite a few, actually) are an artifact that does not effect the code (these plots were meant to be internal to the code...).

CLs with borderline sensitivity

More than one behavior near the edge of the experimental sensitivity is possible. Here are two simple scenarios. In both scenarios, under both hypotheses (i.e. in both ensembles), the LLR is distributed as a Gaussian and the two distributions are displaced by an amount that descreases as we run out of experimental sensitivity.
  • 1 - the Gaussians have the same width
    • in this scenario, the lower the LLR (below the SM peak), the more the SM is prefered.
    • thus, as $\lambda$ increases and we run out of experimental sensitivity, the CLs limit exists, and will rapidly drop to $-\infty$. However, our ability to determine its correct value will detriorate quickly - brute force ensemble testing is not a suitable tool for learning about the tails of distributions.
    • this is the behavior that was observed in the ICHEP results, where increasing the ensemble size by an order of magnitude extended the limit
  • 2 - the new physics distribution is wider
    • in this scenario, for very low LLR values (well below the SM peak), the SM is no longer prefered.
    • thus, as $\lambda$ increases and we run out of experimental sensitivity, the CLs limit no longer exists as the plot is always above 0.05
    • this is the behavior observed in the current results - increasing the ensemble size does not extended the limit

The current borderline sensitivity region

LLR distributions CLs plot  
from 0 to 1 zoomed in  
cpsdn40001I2_88_v2.png ex3.png ex6.png
LLR distributions CLs plot LLR breakdown
from 0 to 1 zoomed in by highest non-zero inner bin
cpsdn40501I2_88_v2.png ex4.png ex5.png more_ex1.png
LLR distributions CLs plot  
from 0 to 1 zoomed in  
cpsdn41001I2_88_v2.png --- ex7.png  

The table

Lambda N_{PDS} At the CLs point Frequentist limiting Data
N_{S+B} CLb CLsb LLR value LLR value LLR value N_{S+B} CLs MC stat. err. on CLs
1.80 4000 199 0.0498 0.9963 -1.01 -0.99 -27.75 0 0.0000 0.0000
1.85 4000 199 0.0497 0.9939 -0.82 -0.78 -25.50 0 0.0000 0.0000
1.90 4000 197 0.0494 0.9875 -1.49 -1.41 -24.14 0 0.0005 0.0005
1.95 4000 198 0.0495 0.9906 -0.78 -0.73 -23.32 1 0.0015 0.0015
2.00 4000 196 0.0489 0.9787 -2.04 -1.92 -21.23 0 0.0002 0.0002
2.20 4000 185 0.0462 0.9238 -2.24 -2.03 -15.19 0 0.0000 0.0000
2.40 4000 177 0.0443 0.8856 -2.17 -1.83 -13.98 0 0.0000 0.0000
2.60 4000 184 0.0460 0.9194 -0.29 -0.12 -10.90 0 0.0000 0.0000
2.80 4000 166 0.0414 0.8276 -0.94 -0.71 -8.84 0 0.0000 0.0000
3.00 4000 84 0.0209 0.4188 -3.10 -1.95 -7.02 2 0.0211 0.0162
3.20 4000 63 0.0157 0.3132 -2.99 -1.81 -5.99 2 0.0181 0.0148
3.40 4000 19 0.0047 0.0933 -3.46 -1.67 -4.86 1 0.0265 0.0268
3.60 8000 --- --- --- --- -1.59 -4.00 7 0.0803 0.0307
3.80 12000 28 0.0023 0.0466 -2.90 -1.38 -3.66 2 0.0261 0.0187
4.00 16000 32 0.0020 0.0395 -3.02 -1.47 -3.57 1 0.0057 0.0057
4.05 12000 --- --- --- --- -1.66 -3.41 6 0.0992 0.0440
4.10 8000 --- --- --- --- -1.57 -3.27 5 0.1350 0.0641
4.15 4000 --- --- --- --- -1.60 -3.18 4 0.2368 0.1259
4.20 4000 --- --- --- --- -1.61 -3.03 4 0.1469 0.0859
5.00 4000 --- --- --- --- -1.07 -1.49 11 0.4184 0.1493

Assorted attachments

Topic attachments
I Attachment History Action Size Date Who Comment
PNGpng cls_wo_syst_5TeV.png r1 manage 8.2 K 2010-09-27 - 21:08 AmnonHarel  
PNGpng cpsdn20001I2_88.png r3 r2 r1 manage 15.9 K 2010-10-07 - 17:03 AmnonHarel  
PNGpng cpsdn20001I2_88_v2.png r1 manage 16.2 K 2010-09-30 - 20:30 AmnonHarel  
PNGpng cpsdn20001I2_88_v4.png r1 manage 16.2 K 2010-10-07 - 16:31 AmnonHarel  
PNGpng cpsdn36001I2_88.png r2 r1 manage 14.8 K 2010-10-07 - 17:03 AmnonHarel  
PNGpng cpsdn36001I2_88_v4.png r1 manage 17.5 K 2010-10-07 - 16:31 AmnonHarel  
PNGpng cpsdn40001I2_88.png r3 r2 r1 manage 15.8 K 2010-10-07 - 17:04 AmnonHarel  
PNGpng cpsdn40001I2_88_v2.png r1 manage 16.7 K 2010-09-30 - 20:31 AmnonHarel  
PNGpng cpsdn40001I2_88_v4.png r1 manage 15.9 K 2010-10-07 - 16:30 AmnonHarel  
PNGpng cpsdn40501I2_88.png r4 r3 r2 r1 manage 15.1 K 2010-10-07 - 17:04 AmnonHarel  
PNGpng cpsdn40501I2_88_v2.png r1 manage 18.0 K 2010-09-30 - 20:33 AmnonHarel  
PNGpng cpsdn40501I2_88_v4.png r1 manage 15.5 K 2010-10-07 - 16:29 AmnonHarel  
PNGpng cpsdn41001I2_88.png r3 r2 r1 manage 14.6 K 2010-10-07 - 17:04 AmnonHarel  
PNGpng cpsdn41001I2_88_v2.png r1 manage 17.9 K 2010-09-30 - 20:31 AmnonHarel  
PNGpng cpsdn41001I2_88_v4.png r1 manage 14.8 K 2010-10-07 - 16:32 AmnonHarel  
PNGpng cpsdn41501I2_88.png r2 r1 manage 14.7 K 2010-10-07 - 17:05 AmnonHarel  
PNGpng cpsdn41501I2_88_v4.png r1 manage 17.7 K 2010-10-07 - 16:29 AmnonHarel  
PNGpng cpsdn50001I2_88.png r2 r1 manage 17.1 K 2010-09-30 - 20:44 AmnonHarel  
PNGpng cpsdn50001I2_88_v2.png r1 manage 17.1 K 2010-09-30 - 20:30 AmnonHarel  
PNGpng cpsdn50001I2_88_v4.png r1 manage 17.1 K 2010-10-07 - 16:30 AmnonHarel  
PNGpng cpsdn50002I2_88.png r1 manage 16.6 K 2010-10-07 - 17:06 AmnonHarel  
PNGpng ex1.png r3 r2 r1 manage 10.9 K 2010-09-30 - 19:00 AmnonHarel  
PNGpng ex2.png r3 r2 r1 manage 12.9 K 2010-09-30 - 19:00 AmnonHarel  
PNGpng ex3.png r3 r2 r1 manage 12.1 K 2010-09-30 - 19:03 AmnonHarel  
PNGpng ex4.png r2 r1 manage 12.7 K 2010-09-30 - 19:01 AmnonHarel  
PNGpng ex5.png r2 r1 manage 14.1 K 2010-09-30 - 19:01 AmnonHarel  
PNGpng ex6.png r2 r1 manage 11.4 K 2010-09-30 - 19:01 AmnonHarel  
PNGpng ex7.png r1 manage 11.3 K 2010-09-30 - 19:02 AmnonHarel  
PNGpng more_ex1.png r1 manage 20.9 K 2010-09-28 - 17:21 AmnonHarel  
PNGpng n8ex1.png r1 manage 10.9 K 2010-09-30 - 19:20 AmnonHarel  
PNGpng n8ex2.png r1 manage 12.9 K 2010-09-30 - 19:20 AmnonHarel  
PNGpng n8ex3.png r1 manage 12.1 K 2010-09-30 - 19:20 AmnonHarel  
PNGpng n8ex4.png r1 manage 12.4 K 2010-09-30 - 19:20 AmnonHarel  
PNGpng n8ex5.png r1 manage 13.9 K 2010-09-30 - 19:21 AmnonHarel  
PNGpng n8ex6.png r1 manage 11.4 K 2010-09-30 - 19:21 AmnonHarel  
PNGpng n8ex7.png r1 manage 10.6 K 2010-09-30 - 19:21 AmnonHarel  
PNGpng n8ex8.png r1 manage 10.6 K 2010-09-30 - 19:35 AmnonHarel  
PNGpng n8ex9.png r1 manage 13.8 K 2010-09-30 - 20:50 AmnonHarel  
PNGpng plot_syst_tail.png r1 manage 21.1 K 2010-10-07 - 12:59 AmnonHarel  
PNGpng v7ex1.png r1 manage 10.9 K 2010-10-07 - 16:54 AmnonHarel  
PNGpng v7ex2.png r1 manage 12.5 K 2010-10-07 - 16:54 AmnonHarel  
PNGpng v7ex5.png r1 manage 13.6 K 2010-10-07 - 16:53 AmnonHarel  
PNGpng v7ex6.png r2 r1 manage 13.5 K 2010-10-07 - 17:15 AmnonHarel  
PNGpng v7ex7.png r2 r1 manage 11.9 K 2010-10-07 - 17:14 AmnonHarel  
PNGpng v7ex8.png r2 r1 manage 11.5 K 2010-10-07 - 17:15 AmnonHarel  
PNGpng v7ex9.png r1 manage 13.5 K 2010-10-07 - 16:50 AmnonHarel  
Edit | Attach | Watch | Print version | History: r13 < r12 < r11 < r10 < r9 | Backlinks | Raw View | Raw edit | More topic actions...
Topic revision: r10 - 2010-10-07 - AmnonHarel
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    Main All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback