1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS...
-
Upload
melanie-hurst -
Category
Documents
-
view
214 -
download
0
Transcript of 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS...
![Page 1: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/1.jpg)
1
Spatial processes and statistical modelling
Peter GreenUniversity of Bristol, UK
BCCS GM&CSS 2008/09 Lecture 8
![Page 2: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/2.jpg)
2
• Continuous space
• Discrete space– lattice– irregular - general graphs– areally aggregated
• Point processes– other object processes
Spatial indexing
![Page 3: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/3.jpg)
3
Space vs. time
• apparently slight difference• profound implications for
mathematical formulation and computational tractability
![Page 4: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/4.jpg)
4
Requirements of particular application domains• agriculture (design)
• ecology (sparse point pattern, poor data?)
• environmetrics (space/time)
• climatology (huge physical models)
• epidemiology (multiple indexing)
• image analysis (huge size)
![Page 5: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/5.jpg)
5
Key themes
• conditional independence– graphical/hierarchical modelling
• aggregation– analysing dependence between differently
indexed data– opportunities and obstacles
• literal credibility of models• Bayes/non-Bayes distinction blurred
![Page 6: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/6.jpg)
6
Why build spatial dependence into a model?• No more reason to suppose
independence in spatially-indexed data than in a time-series
• However, substantive basis for form of spatial dependent sometimes slight - very often space is a surrogate for missing covariates that are correlated with location
![Page 7: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/7.jpg)
7
Discretely indexed data
![Page 8: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/8.jpg)
8
Modelling spatial dependence in discretely-indexed fields• Direct• Indirect
– Hidden Markov models– Hierarchical models
![Page 9: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/9.jpg)
9
Hierarchical models, using DAGs
Variables at several levels - allows modelling of complex systems, borrowing strength, etc.
![Page 10: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/10.jpg)
10
Modelling with undirected graphsDirected acyclic graphs are a natural
representation of the way we usually specify a statistical model - directionally:
• disease symptom• past future• parameters data ……whether or not causality is understood.But sometimes (e.g. spatial models) there is
no natural direction
![Page 11: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/11.jpg)
11
Conditional independence
In model specification, spatial context often rules out directional dependence (that would have been acceptable in time series context)
X0 X1 X2 X3 X4
![Page 12: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/12.jpg)
12
Conditional independence
In model specification, spatial context often rules out directional dependence
X20 X21 X22 X23 X24
X00 X01 X02 X03 X04
X10 X11 X12 X13 X14
![Page 13: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/13.jpg)
13
Conditional independence
In model specification, spatial context often rules out directional dependence
X20 X21 X22 X23 X24
X00 X01 X02 X03 X04
X10 X11 X12 X13 X14
![Page 14: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/14.jpg)
14
Directed acyclic graph
)|()( )(pa vVv
v xxpxp
in general:
for example:
a b
c
dp(a,b,c,d)=p(a)p(b)p(c|a,b)p(d|c)In the RHS, any distributions are legal, and uniquely define joint distribution
![Page 15: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/15.jpg)
15
Undirected (CI) graph
X20 X21 X22
X00 X01 X02
X10 X11 X12Absence of edge denotes conditional independence given all other variables
But now there are non-trivial constraints on conditional distributions
Regular lattice, irregular graph, areal data...
![Page 16: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/16.jpg)
16
Undirected (CI) graph
X20 X21 X22
X00 X01 X02
X10 X11 X12
C
CC XVXp ))(exp)(
iC
CCii XVXXp ))(exp)|(
)|()|( iiii XXpXXp
then
and so
The Hammersley-Clifford theorem says essentially that the converse is also true - the only sure way to get a valid joint distribution is to use ()
()
clique
Suppose we assume
![Page 17: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/17.jpg)
17
Hammersley-Clifford
X20 X21 X22
X00 X01 X02
X10 X11 X12
C
CC XVXp )(exp)(
)|()|( iiii XXpXXp
A positive distribution p(X) is a Markov random field
if and only if it is a Gibbs distribution
- Sum over cliques C (complete subgraphs)
![Page 18: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/18.jpg)
18
Partition function
X20 X21 X22
X00 X01 X02
X10 X11 X12
C
CC XVXp )(exp)(
Almost always, the constant of proportionality in
is not available in tractable form: an obstacle to likelihood or Bayesian inference about parameters in the potential functions
Physicists call
the partition function
))( CC XV
X C
CC XVZ )(exp
![Page 19: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/19.jpg)
19
Markov properties for undirected graphs• The situation is a bit more complicated
than it is for DAGs. There are 4 kinds of Markovness:
• P – pairwise– Non-adjacent pairs of variables are
conditionally independent given the rest
• L – local– Conditional only on adjacent variables
(neighbours), each variable is independent of all others
![Page 20: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/20.jpg)
20
• G – global– Any two subsets of variables separated by a
third are conditionally independent given the values of the third subset.
• F – factorisation– the joint distribution factorises as a product
of functions of cliques• In general these are different, but FGLP
always. For a positive distribution, they are all the same.
![Page 21: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/21.jpg)
21
Gaussian Markov random fields: spatial autoregression
C
CC XVXp )(exp)(
)|()|( iiii XXpXXp is a multivariate Gaussian distribution, and
If VC(XC) is -ij(xi-xj)2/2 for C={i,j} and 0 otherwise, then
is the univariate Gaussian distribution
),(~| 22i
ijjijiii XNXX
ij
iji /12where
![Page 22: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/22.jpg)
22
Inverse of (co)variance matrix:
dependent case3210
2410
1121
0012
A B C D
non-zero
non-zero),|,cov( CADB
A B C D
A
B
C
D
Gaussian random fields
![Page 23: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/23.jpg)
23
Non-Gaussian Markov random fields
ji
ji xx
~
)cosh(log)1(exp
C
CC XVXp )(exp)(
Pairwise interaction random fields with less smooth realisations obtained by replacing squared differences by a term with smaller tails, e.g.
![Page 24: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/24.jpg)
24
Discrete-valued Markov random fields
C
CC XVXp )(exp)(
Besag (1974) introduced various cases of
for discrete variables, e.g. auto-logistic (binary variables), auto-Poisson (local conditionals are Poisson), auto-binomial, etc.
![Page 25: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/25.jpg)
25
Auto-logistic model
C
CC XVXp )(exp)(
ji
jiiji
ii xxx~
exp
)|()|( iiii XXpXXp is Bernoulli(pi) with
ij
jijiii xpp )())1/(log(
- a very useful model for dependent binary variables (NB various parameterisations)
(Xi = 0 or 1)
![Page 26: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/26.jpg)
26
Statistical mechanics models
C
CC XVXp )(exp)(
The classic Ising model (for ferromagnetism) is the symmetric autologistic model on a square lattice in 2-D or 3-D. The Potts model is the generalisation to more than 2 ‘colours’
ji
jii
x xxIXpi
~
][exp)(
and of course you can usefully un-symmetrise this.
![Page 27: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/27.jpg)
27
Auto-Poisson model
C
CC XVXp )(exp)(
ji
jiiji
iii xxxx~
)!log(exp
)|()|( iiii XXpXXp is Poisson ))(exp(
ij
jiji x
For integrability, ij must be 0, so this onlymodels negative dependence: very limited use.
![Page 28: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/28.jpg)
28
Hierarchical models and hidden Markov processes
![Page 29: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/29.jpg)
29
Chain graphs
• If both directed and undirected edges, but no directed loops:
• can rearrange to form global DAG with undirected edges within blocks
![Page 30: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/30.jpg)
30
Chain graphs
• If both directed and undirected edges, but no directed loops:
• can rearrange to form global DAG with undirected edges within blocks
• Hammersley-Clifford within blocks
![Page 31: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/31.jpg)
31
Hidden Markov random fields• We have a lot of freedom modelling
spatially-dependent continuously-distributed random fields on regular or irregular graphs
• But very little freedom with discretely distributed variables
use hidden random fields, continuous or discrete
• compatible with introducing covariates, etc.
![Page 32: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/32.jpg)
32
Hidden Markov models
z0 z1 z2 z3 z4
y1 y2 y3 y4
e.g. Hidden Markov chain
observed
hidden
![Page 33: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/33.jpg)
33
Hidden Markov random fields
Unobserved dependent field
Observed conditionally-independent discrete field
(a chain graph)
![Page 34: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/34.jpg)
34
Spatial epidemiology applications
independently, for each region i. Options:• CAR, CAR+white noise (BYM, 1989)• Direct modelling of ,e.g. SAR• Mixture/allocation/partition models:
• Covariates, e.g.:
)(~ iii ePoissonY casesexpected
cases
relative risk
)log( i))cov(log( i
izi eYEi
)(
)exp()( Tiizi xeYE
i
![Page 35: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/35.jpg)
35
Spatial epidemiology applications
Spatial contiguity is usually somewhat idealised
![Page 36: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/36.jpg)
36
relativerisk
parameters
Richardson & Green (JASA, 2002) used a hidden Markov random field model for disease mapping
)(Poisson~ izi eyi
observedincidence
expectedincidencehidden
MRF
Spatial epidemiology applications
![Page 37: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/37.jpg)
37)(Poisson~ ii eY
Chain graph for disease mapping
e
Y
zk
)(Poisson~ izi eYi
based on Potts model
![Page 38: 1 Spatial processes and statistical modelling Peter Green University of Bristol, UK BCCS GM&CSS 2008/09 Lecture 8.](https://reader035.fdocuments.us/reader035/viewer/2022070306/5515de4a550346dd6f8b4bdf/html5/thumbnails/38.jpg)
38
Larynx cancer in females in France
SMRs
)|1( ypiz
ii Ey /