1. 1 Practical Statistics for Physicists Stockholm Lectures Sept 2008 Louis Lyons Imperial College and Oxford CDF experiment at FNAL CMS expt at LHC l.lyons@physics.ox.ac.uk

2. 2 Topics 1) Introduction Learning to love the Error Matrix 2) Do’s and Dont’s with L ikelihoods χ 2 and Goodness of Fit 3) Discovery and p-values 4) Bayes and Frequentism Plenty of time for discussion

3. 3 Some of the questions to be addressed What is coverage, and do we really need it? Should we insist on at least a 5σ effect to claim discovery? How should p-values be combined? If two different models both have respectable χ 2 probabilities, can we reject one in favour of other? Are there different possibilities for quoting the sensitivity of a search? How do upper limits change as the number of observed events becomes smaller than the predicted background? Combine 1 ± 10 and 3 ± 10 to obtain a result of 6 ± 1? What is the Punzi effect and how can it be understood?

4. 4 Books Statistics for Nuclear and Particle Physicists Cambridge University Press, 1986 Available from CUP Errata in these lectures

5. 5 Other Books CDF Statistics Committee BaBar Statistics Working Group

6. 6

7. 7 Introductory remarks Probability and Statistics Random and systematic errors Conditional probability Variance Combining errors Combining experiments Binomial, Poisson and Gaussian distributions

8. 8

9. 9 Parameter Determination Goodness of fit Hypothesis testing THEORY  DATA DATA  THEORY N.B. Parameter values not sensible if goodness of fit is poor/bad

10. 10 Why do we need errors? Affects conclusion about our result e.g. Result / theory = 0.970 If 0.970 ± 0.050, data compatible with theory If 0.970 ± 0.005, data incompatible with theory If 0.970 ± 0.7, need better experiment Historical experiment at Harwell testing General Relativity

11. 11 Random + Systematic Errors Random/Statistical: Limited accuracy, Poisson counts Spread of answers on repetition (Method of estimating) Systematics: May cause shift, but not spread e.g. Pendulum g = 4π 2 L/τ, τ = T/n Statistical errors: T, L Systematics: T, L Calibrate: Systematic  Statistical More systematics: Formula for undamped, small amplitude, rigid, simple pendulum Might want to correct to g at sea level: Different correction formulae Ratio of g at different locations: Possible systematics might cancel. Correlations relevant

12. 12 Presenting result Quote result as g ± σ stat ± σ syst Or combine errors in quadrature  g ± σ Other extreme: Show all systematic contributions separately Useful for assessing correlations with other measurements Needed for using: improved outside information, combining results using measurements to calculate something else.

13. 13

14. 14

15. 15 Combining errors z = x - y δz = δx – δy [1] Why σ z 2 = σ x 2 + σ y 2 ? [2]

16. 16 Combining errors z = x - y δz = δx – δy [1] Why σ z 2 = σ x 2 + σ y 2 ? [2] 1)[1] is for specific δx, δy Could be so on average ? N.B. Mneumonic, not proof 2) σ z 2 = δz 2 = δx 2 + δy 2 – 2 δx δy = σ x 2 + σ y 2 provided…………..

17. 17 3) Averaging is good for you: N measurements x i ± σ [1] x i ± σ or [2] x i ± σ/ √N ? 4) Tossing a coin: Score 0 for tails, 2 for heads (1 ± 1 ) After 100 tosses, [1] 100 ± 100 or [2] 100 ± 10 ? 0 100 200 Prob(0 or 200) = (1/2) 99 ~ 10 -30 Compare age of Universe ~ 10 18 seconds

18. 18 Rules for different functions 1)Linear: z = k 1 x 1 + k 2 x 2 + ……. σ z = k 1 σ 1 & k 2 σ 2 & means “combine in quadrature” 2) Products and quotients z = x α y β ……. σ z /z = α σ x /x & β σ y /y

19. 19 3) Anything else: z = z(x 1, x 2, …..) σ z = ∂z/∂x 1 σ 1 & ∂z/∂x 2 σ 2 & ……. OR numerically: z 0 = z(x 1, x 2, x 3 ….) Z 1 = f(x 1 +σ 1, x 2, x 3 ….) Z 2 = f(x 1, x 2 + σ 2, x 3 ….) σ z = (z 1 -z 0 ) & (z 2 -z 0 ) & ….

20. 20

21. 21 To consider…… Is it possible to combine 1 ± 10 and 2 ± 9 to get a best combined value of 6 ± 1 ? Answer later.

22. 22 Difference between averaging and adding Isolated island with conservative inhabitants How many married people ? Number of married men = 100 ± 5 K Number of married women = 80 ± 30 K Total = 180 ± 30 K Wtd average = 99 ± 5 K CONTRAST Total = 198 ± 10 K GENERAL POINT: Adding (uncontroversial) theoretical input can improve precision of answer Compare “kinematic fitting”

23. 23 Binomial Distribution Fixed N independent trials, each with same prob of success p What is prob of s successes? e.g. Throw dice 100 times. Success = ‘6’. What is prob of 0, 1,…. 49, 50, 51,… 99, 100 successes? Effic of track reconstrn = 98%. For 500 tracks, prob that 490, 491,...... 499, 500 reconstructed. Ang dist is 1 + 0.7 cos θ? Prob of 52/70 events with cosθ > 0 ? (More interesting is statistics question)

24. 24 P s = N! p s (1-p) N-s, as is obvious (N-s)! s! Expected number of successes = ΣnP n = Np, as is obvious Variance of no. of successes = Np(1-p) Variance ~ Np, for p~0 ~ N(1-p) for p~1 NOT Np in general. NOT n ±√n

25. 25 Statistics: Estimate p and σ p from s (and N) p = s/N σ p 2 = 1/N s/N (1 – s/N) If s = 0, p = 0 ± 0 ? If s = 1, p = 1.0 ± 0 ? Limiting cases: ● p = const, N  ∞: Binomial  Gaussian μ = Np, σ 2 = Np(1-p) ● N  ∞, p  0, Np = const: Bin  Poisson μ = Np, σ 2 = Np {N.B. Gaussian continuous and extends to -∞}

26. 26

27. 27 Poisson Distribution Prob of n independent events occurring in time t when rate is r (constant) e.g. events in bin of histogram NOT Radioactive decay for t ~ τ Limit of Binomial (N  ∞, p  0, Np  μ) P n = e -r t (r t) n /n! = e - μ μ n /n! (μ = r t) = r t = μ (No surprise!) σ 2 n = μ “n ±√n” BEWARE 0 ± 0 ? μ  ∞: Poisson  Gaussian, with mean = μ, variance =μ Important for χ 2

28. 28 For your thought Poisson P n = e - μ μ n /n! P 0 = e – μ P 1 = μ e –μ P 2 = μ 2 /2 e -μ For small μ, P 1 ~ μ, P 2 ~ μ 2 /2 If probability of 1 rare event ~ μ, Why isn’t probability of 2 events ~ μ 2 ?

29. 29

30. 30 Relation between Poisson and Binomial PeopleMaleFemale PatientsCuredRemain ill Decaying nuclei ForwardsBackwards Cosmic raysProtonsOther particles N people in lecture, m males and f females (N = m + f ) Assume these are representative of basic rates : ν people νp males ν(1-p) females Probability of observing N people = P Poisson = e –ν ν N /N! Prob of given male/female division = P Binom = p m (1-p) f Prob of N people, m male and f female = P Poisson P Binom = e –νp ν m p m * e -ν(1-p) ν f (1-p) f m! f ! = Poisson prob for males * Poisson prob for females

31. 31 Gaussian or Normal

32. 32

33. 33 Relevant for Goodness of Fit

34. 34

35. 35 Gaussian = N (r, 0, 1) Breit Wigner = 1/{π * (r 2 + 1)}

36. 36

37. 37 Learning to love the Error Matrix Introduction via 2-D Gaussian Understanding covariance Using the error matrix Combining correlated measurements Estimating the error matrix

38. 38 Element E ij - Diagonal E ij = variances Off-diagonal E ij = covariances

39. 39

40. 40

41. 41

42. 42

43. 43

44. 44

45. 45

46. 46

47. 47

48. 48

49. 49

50. 50 Small error Example: Lecture 2 x best outside x 1  x 2 y best outside y 1  y 2

51. 51 a b x y

52. 52

53. 53

54. 54

55. 55 Conclusion Error matrix formalism makes life easy when correlations are relevant

56. 56 Next time: L ikelihoods What it is How it works: Resonance Error estimates Detailed example: Lifetime Several Parameters Extended maximum L Do’s and Dont’s with L 