Wing Sheung Chan

Statistical interpretation and results 119 Table 6.4.: Best-fit values and uncertainties of B ( Z → µτ ) and the normalisation factors, and the expected (median) and observed upper limits on B ( Z → µτ ) at 95% CL. Results from standalone fits to the Run 1 or Run 2 data set are shown alongside the results from the combined fit for comparison. The uncertainties include both the statistical and systematic contributions. Run 1 Run 2 Combined B ( Z → µτ ) [ 10 − 6 ] − 15 . 8 +12 . 1 − 11 . 5 4 . 3 ± 3 . 2 3 . 1 ± 3 . 1 µ Z (Run 2) – 0 . 97 ± 0 . 02 0 . 97 ± 0 . 02 µ 1P fakes (Run 2) – 1 . 00 ± 0 . 02 1 . 01 ± 0 . 02 µ 3P fakes (Run 2) – 1 . 14 ± 0 . 04 1 . 14 ± 0 . 04 µ Z (Run 1) 1 . 02 +0 . 07 − 0 . 06 – 1 . 02 +0 . 07 − 0 . 06 µ W (Run 1) 1 . 10 +0 . 09 − 0 . 08 – 1 . 09 ± 0 . 07 Expected upper limit at 95% CL [ 10 − 6 ] 25.7 6.3 6.1 Observed upper limit at 95% CL [ 10 − 6 ] 16.9 9.9 9.5 statistical and systematic uncertainties related to fakes estimation are also uncorrelated in the two analyses. Most of the object reconstruction and identification algorithms have been significantly changed in Run 2 compared to Run 1. This is especially true for τ had - vis algorithms – the MVA TES, RNN ID and e -veto BDT are all newly developed algorithms that differ from the algorithms used in Run 1. Performance measurements, which are based on data, are also statistically uncorrelated in the two runs. In view of all the above, it appears reasonable to consider all the NPs in the two fit models to be uncorrelated. The best-fit values and uncertainties of the POI and NFs after a fit to the Run 1 and Run 2 combined data set, as well as the expected and observed upper limits at 95% CL, are shown in Table 6.4, alongside the results from standalone fits to the Run 1 or Run 2 data set for comparison. The observed upper limit on B ( Z → µτ ) from the combined measurement is 9 . 5 × 10 − 6 .