Modeling Dependence in Uncertainty Analysis - PowerPoint PPT Presentation

1 / 15
About This Presentation
Title:

Modeling Dependence in Uncertainty Analysis

Description:

Diagonal Band Min Information Elliptical Copula. r(bldpr,út)=90%; r(bldpr, wgt)=80 ... Elliptical Copula. Partial correlation = Conditional correlation ... – PowerPoint PPT presentation

Number of Views:90
Avg rating:3.0/5.0
Slides: 16
Provided by: rogerm94
Category:

less

Transcript and Presenter's Notes

Title: Modeling Dependence in Uncertainty Analysis


1
Modeling Dependence in Uncertainty Analysis
  • Roger M. Cooke
  • Inst. Appl. Mathematics
  • Delft University of Technology
  • r.m.cooke_at_its.tudelft.nl
  • http//ssor.twi.tudelft.nl/risk/.

2
How to Represent Dependence
  • Via functional relations
  • Via dependence trees
  • Via vines

3
Functional relations
  • XT X1Xn independent standardized
  • C ? COV(n x n) C LLT L lower triangular
  • Y LX has covariance matrix C.
  • (what are the marginals Y1,Yn????)

4
Dependence trees
  • Marginals are given, how characterize dependence?

bloodpressure
Fat
weight
5
Rank Correlation
  • Copula bivariate uniform with (rank) correlation
  • (i.e. correlation between percentiles)
  • Diagonal Band Min Information Elliptical Copula

6
r(bldpr,Fat)90 r(bldpr, wgt)80Cobweb
Plots, (min inf copua)natural scale
Percentile scale
7
r(bldpr,Fat)90 r(bldpr, wgt)80Cobweb
Plots, (Elliptical copua)natural scale
Percentile scale
8
More Dependences
  • With n vbls, there are n-choose-2 correlations
  • There are n-1 edges in a tree

9
Belief Nets (directed acyclic graph)
  • Cant simply assign correlations to arcs wont
    be positive definite

10
Elliptical Copula
  • Partial correlation Conditional correlation
  • Conditional correlations dont depend on value of
    conditioning vbls
  • Partial correlations on a Regular Vine uniquely
    determine correlation matrix
  • Partial correlations on a Regular Vine are
    algebraically independent
  • Replace (conditional) independence with
    (conditional) indifference (correlation zero)

11
Associate (conditional) rank correlations
r(B,F W,H)
r(B,W)
r(B,H W)
r(W,F)
r(W,H F)
12
Regular (D-) Vine Nested set of treesEdges are
joined only if they have common child
r(F,H)
r(W,F)
r(B,W)
Weight
Bldprssr
Fat
Height
r(B,F W)
r(W,H F)
r(B,H W,F)
13
Assess conditional rank correlations
  • r(B,F W)
  • Suppose an individual is chosen with weight
    60kg, suppose for him/her the fat in diet is
    above the median value for 60kg people, what is
    your probability that for him/her the blood
    pressure is also above the median?

14
Unconditional CobwebsVine
Tree
15
Conditional Cobwebs (top 10 bldprssr)Vine
Tree
Write a Comment
User Comments (0)
About PowerShow.com