A Cantelli-type inequality for constructing non-parametric p-boxes - - PowerPoint PPT Presentation

a cantelli type inequality for constructing non
SMART_READER_LITE
LIVE PREVIEW

A Cantelli-type inequality for constructing non-parametric p-boxes - - PowerPoint PPT Presentation

A Cantelli-type inequality for constructing non-parametric p-boxes based on exchangeability Matthias C. M. Troffaes Tathagata Basu Department of Mathematical Sciences Durham University, UK July 2019 Outline 1 Cantellis inequality and


slide-1
SLIDE 1

A Cantelli-type inequality for constructing non-parametric p-boxes based on exchangeability

Matthias C. M. Troffaes Tathagata Basu

Department of Mathematical Sciences Durham University, UK

July 2019

slide-2
SLIDE 2

Outline

1 Cantelli’s inequality and p-boxes 2 Contributions 3 Conclusions

slide-3
SLIDE 3

Cantelli’s inequality & induced p-box

  • random variable X, known mean µ and variance σ2
  • Cantelli’s inequality [1]:

0 ≤ P X − µ σ ≤ λ

1 1 + λ2 if λ ≤ 0, (1a) λ2 1 + λ2 ≤ P X − µ σ ≤ λ

  • ≤ 1

if λ ≤ 0. (1b)

  • induces a p-box (lower & upper cdf, bounding a set of

probability measures) Issue What if only sample mean and sample standard deviation are known?

slide-4
SLIDE 4

Contributions: Problem Statement

Assumptions X1,. . . ,Xn, Xn+1 is a finite sequence of discrete exchangeable random variables. Notation

  • X := 1

n

n

j=1 Xj

  • S2 :=

n

j=1(Xj−X)2

(n−1)

Objective Find functions f and f such that f (λ, n) ≤ P Xn+1 − X S ≤ λ

  • ≤ f (λ, n)

(2)

slide-5
SLIDE 5

Contributions: Cantelli-type inequality

Theorem For every λ ≥ 0, 1 n + 1 (n + 1)λ2

n

λ2

n + 1

  • ≤ P
  • Xn+1 − X

S + ∆n

√n

< λ

  • ≤ 1

(3) where λn :=

n √ n2−1λ and ∆n :=

  • n+1

n−1(max Xj − min Xj).

Similarly, for every λ ≤ 0, 0 ≤ P

  • Xn+1 − X

S + ∆n

√n

≤ λ

1 n + 1 n + 1 λ2

n + 1

  • .

(4) Here, ⌊x⌋ := max{n ∈ Z: n ≤ x} and ⌈x⌉ := −⌊−x⌋.

slide-6
SLIDE 6

Contributions: Cantelli-type inequality

Plotting left and right hand sides from inequalities in theorem:

slide-7
SLIDE 7

Contributions: P-box and Prediction Interval

Inequalities from theorem induce the following: Non-parametric p-box . . . on the random variable Zn+1 = Xn+1−X

S+ ∆n √n

. Not a p-box on Xn+1 directly! Prediction interval . . . on the random variable Xn+1. For any ℓ1 and ℓ2, we can calculate α1 and α2 such that α1 ≤ P(X − ℓ1Sn < Xn+1 ≤ X + ℓ2Sn) ≤ α2 (5) where, Sn := S + ∆n

√n.

slide-8
SLIDE 8

Conclusions I

  • novel Cantelli-type inequality
  • induces non-parametric p-box and prediction interval
  • only assumes exchangeability (rather than conditional

independence)

  • only uses sample mean and sample standard deviation
  • similar to Saw [2] (but Saw does not induce a p-box)
  • useful for modelling when only sample mean and sample

standard deviation are known (e.g. measurement problems)

slide-9
SLIDE 9

Conclusions II

  • p-box only on Zn+1 and not on Xn+1
  • use of prediction interval not entirely clear
  • p-box on Xn+1 , conditional on X and S, using

exchangeability, remains and open problem (might be impossible as pointed out by a kind reviewer)

slide-10
SLIDE 10

Thank you for your attention! We look forward to seeing you at our poster!

slide-11
SLIDE 11

References

  • F. P. Cantelli. “Sui confini della probabilit`

a”. In: Atti del Congresso Internazional del Matematici 6 (1928). Bologna,

  • pp. 47–59.

John G. Saw, Mark C. K. Yang, and Tse Chin Mo. “Chebyshev Inequality with Estimated Mean and Variance”. In: The American Statistician 38.2 (1984), pp. 130–132.