An Infinitesimal Version of Stein’s Method of Exchangeable ...
Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf ·...
Transcript of Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf ·...
![Page 1: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/1.jpg)
A Short Introduction to Stein’s Method
Gesine Reinert
Department of Statistics
University of Oxford
1
![Page 2: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/2.jpg)
Overview
Lecture 1: focusses mainly on normal approximation
Lecture 2: other approximations
2
![Page 3: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/3.jpg)
1. The need for bounds
Distributional approximations:
Example X1, X2, . . . , Xn i.i.d., P(Xi = 1) = p = 1 −
P(Xi = 0)
n−1/2 n∑i=1
(Xi − p) ≈d N (0, p(1− p))
n∑i=1
Xi ≈d Poisson(np)
would like to assess distance of distributions; would like
bounds
3
![Page 4: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/4.jpg)
Weak convergence
For c.d.f.s Fn, n ≥ 0 and F on the line we say that Fn
converges weakly (converges in distribution) to F ,
Fnw−→ F
if
Fn(x) → F (x) (n →∞)
for all continuity points x of F
For the associated probability distributions:
Pnw−→ P
4
![Page 5: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/5.jpg)
Facts:
1.
Pnw−→ P ⇐⇒ Pn(A) → P (A)
for each P -continuity set A (i.e. P (∂A) = 0)
2.
Pnw−→ P ⇐⇒
∫fdPn →
∫fdP
for all functions f that are bounded, continuous, real-
valued
5
![Page 6: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/6.jpg)
3.
Pnw−→ P ⇐⇒
∫fdPn →
∫fdP
for all functions f that are bounded, infinitely often
differentiable, continuous, real-valued
4. If X is a random variable, denote its distribution by
L(X). Then
L(Xn)w−→ L(X) ⇐⇒ Ef (Xn) → Ef (X)
for all functions f that are bounded, infinitely often
differentiable, continuous, real-valued
6
![Page 7: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/7.jpg)
Metrics
Let L(X) = P,L(Y ) = Q; define total variation dis-
tance
dTV (P, Q) = supA measurable
|P (A)−Q(A)|
7
![Page 8: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/8.jpg)
Put
L = {g : R → R; |g(y)− g(x)| ≤ |y − x|}
and Wasserstein distance
dW (P, Q) = supg∈L
|Eg(Y )− Eg(X)|
= inf E|Y −X|,
where the infimum is over all couplings X,Y such that
L(X) = P,L(Y ) = Q
8
![Page 9: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/9.jpg)
Using
F = {f ∈ L absolutely continuous, f(0) = f ′(0) = 0}
we also have
dW (P, Q) = supf∈F
|Ef ′(Y )− Ef ′(X)|.
9
![Page 10: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/10.jpg)
2. Stein’s Method for Normal Approxima-
tion
Stein (1972, 1986)
Z ∼ N (µ, σ2) if and only if for all smooth functions f ,
E(Z − µ)f (Z) = σ2Ef ′(Z)
For W with EW = µ, VarW = σ2, if
σ2Ef ′(W )− E(W − µ)f (W )
is close to zero for many functions f , then W should be
close to Z in distribution
10
![Page 11: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/11.jpg)
Sketch of proof for µ = 0, σ2 = 1:
Assume Z ∼ N (0, 1). Integration by parts:
1√2π
∫f ′(x)e−x2/2dx
= 1√
2πf (x)e−x2/2
+1√2π
∫xf (x)e−x2/2dx
=1√2π
∫xf (x)e−x2/2dx
Assume EZf (Z) = Ef ′(Z): Can use partial integra-
tion to solve differential equation
f ′(x)− xf (x) = g(x), limx→−∞f (x)e−x2/2 = 0
11
![Page 12: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/12.jpg)
for any bounded function g, giving
f (y) = ey2/2∫ y−∞ g(x)e−x2/2dx
Take g(x) = 1(x ≤ x0)− Φ(x0), then
0 = E(f ′(Z)− Zf (Z)) = P(Z ≤ x0)− Φ(x0)
so Z ∼ N (0, 1).
12
![Page 13: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/13.jpg)
Let µ = 0. Given a test function h, let Nh = Eh(Z/σ),
and solve for f in the Stein equation
σ2f ′(w)− wf (w) = h(w/σ)−Nh
giving
f (y) = ey2/2∫ y−∞ (h(x/σ)−Nh) e−x2/2dx
Now evaluate the expectation of the r.h.s. of the Stein
equation by the expectation of the l.h.s.
Can bound, e.g. ‖ f ′′ ‖≤ 2 ‖ h′ ‖
13
![Page 14: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/14.jpg)
Example: X,X1, . . . , Xn i.i.d. mean zero, VarX = 1n
W =n∑
i=1Xi
Put
Wi = W −Xi =∑j 6=i
Xj
Then
EWf (W ) =n∑
i=1EXif (W )
=n∑
i=1EXif (Wi) +
n∑i=1
EX2i f
′(Wi) + R
=1
n
n∑i=1
Ef ′(Wi) + R
14
![Page 15: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/15.jpg)
So
Ef ′(W )− EWf (W ) =1
n
n∑i=1
E{f ′(W )− f ′(Wi)}
+R
and can bound remainder term R;
Theorem 1 For any smooth h
|Eh(W )−Nh| ≤ ‖h′‖ 2√
n+
n∑i=1
E|X3i | .
Extends to local dependence:
15
![Page 16: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/16.jpg)
Let X1, . . . , Xn be mean zero, finite variances, put
W =n∑
i=1Xi
Assume V arW = 1. Suppose that for each i = 1, . . . , n
there exist sets Ai ⊂ Bi ⊂ {1, . . . , n} such that
Xi is independent of ∑j 6∈Ai
Xj and
∑j∈Ai
Xj is independent of ∑j 6∈Bi
Xj
Define
ηi =∑
j∈Ai
Xj
τi =∑
j 6∈Bi
Xj
16
![Page 17: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/17.jpg)
Theorem 2 For any smooth h with ‖h′‖ ≤ 1,
|Eh(W )−Nh| ≤ 2n∑
i=1(E|Xiηiτi| + |E(Xiηi)|E|τi|)
+n∑
i=1E|Xiη
2i |.
17
![Page 18: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/18.jpg)
Example: Graphical dependence
V = {1, . . . , n} set of vertices in graph G = (V, E)
G is a dependency graph if, for any pair of disjoint sets
Γ1 and Γ2 of V such that no edge in E has one endpoint
in Γ1 and the other endpoint in Γ2, the sets of random
variables {Xi, i ∈ Γ1} and {Xi, i ∈ Γ2} are independent
Let Ai be the set of all j such that (i, j) ∈ E, union
with {i}, Bi = ∪j∈AiAj. Then the above theorem ap-
plies.
18
![Page 19: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/19.jpg)
Size-Bias coupling: µ > 0
If W ≥ 0,EW > 0 then W s has the W -size biased
distribution if
EWf (W ) = EWEf (W s)
for all f for which both sides exist
19
![Page 20: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/20.jpg)
Put f (x) = 1(x = k) then
kP(W = k) = EWP(W s = k)
so
P(W s = k) =kP(W = k)
EW
Example: If X ∼ Bernoulli(p), then EXf (X) =
pf (1) and so Xs = 1
Example: If X ∼ Poisson(λ), then
P(Xs = k) =ke−λλk
k!λ=
e−λλk−1
(k − 1)!
and so Xs = X + 1, where the equality is in distribution
20
![Page 21: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/21.jpg)
Construction
(Goldstein + Rinott 1996) Suppose W = ∑ni=1 Xi
with Xi ≥ 0, EXi > 0, all i.
Choose index V proportional to the mean, EXv. If
V = v: replace Xv by Xsv having the Xv-size biased
distribution, independent, and if Xsv = x: adjust Xu, u 6=
v, such that
L(Xu, u 6= v) = L(Xu, u 6= v|Xv = x)
Then W s = ∑u 6=V Xu + Xs
V
21
![Page 22: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/22.jpg)
Example: Xi ∼ Be(pi) for i = 1, . . . , n
Then W s = ∑u 6=V Xu + 1
See Poisson approximation, Barbour, Holst, Janson
1992
22
![Page 23: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/23.jpg)
X,X1, . . . , Xn ≥ 0 i.i.d., EX = µ, VarX = σ2
W =n∑
i=1Xi
Then
E(W − µ)f (W ) = µE(f (W s)− f (W ))
≈ µE(W s −W )f ′(W )
= µ1
n
n∑i=1
E(Xsi −Xi)f
′(W )
≈ µEf ′(W )E(Xs − µ)
= µEf ′(W )
1
µEX2 − µ
= σ2Ef ′(W ).
23
![Page 24: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/24.jpg)
Zero bias coupling
Let X be a mean zero random variable with finite,
nonzero variance σ2. We say that X∗ has the X-zero
biased distribution if for all differentiable f for which
EXf (X) exists,
EXf (X) = σ2Ef ′(X∗).
The zero bias distribution X∗ exists for all X that have
mean zero and finite variance. (Goldstein and R. 1997)
24
![Page 25: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/25.jpg)
It is easy to verify that W ∗ has density
p∗(w) = σ−2E{W I(W > w)}
Example: If X ∼ Bernoulli(p)− p, then
E{XI(X > x)} = p(1− p) for − p < x < 1− p
and is zero elsewhere, so X∗ ∼ Uniform(−p, 1− p)
25
![Page 26: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/26.jpg)
Connection with Wasserstein distance
We have for W mean zero, variance 1,
|Eh(W )−Nh| = |E [f ′(W )−Wf (W )] |
= |E [f ′(W )− f ′(W ∗)] |
≤ ||f ′′||E|W −W ∗|,
where || · || is the supremum norm. As ||f ′′|| ≤ 2||h′||
|Eh(W )−Nh| ≤ 2||h′||E|W −W ∗|;
thus
dW (L(W ),N (0, 1)) ≤ 2E|W −W ∗|.
26
![Page 27: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/27.jpg)
Construction in the case of W = ∑ni=1 Xi sum of inde-
pendent mean zero finite variance σ2i variables: Choose
an index I proportional to the variance, zero bias in that
variable,
W ∗ = W −XI + X∗I .
27
![Page 28: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/28.jpg)
Proof: For any smooth f ,
EWf (W ) =n∑
i=1EXif (W )
=n∑
i=1EXif (Xi +
∑t6=i
Xt)
=n∑
i=1σ2
i Ef ′(X∗i +
∑t6=i
Xt)
= σ2 n∑i=1
σ2i
σ2Ef ′(W −Xi + X∗
i )
= σ2Ef ′(W −XI + X∗I ) = σ2Ef ′(W ∗),
where we have used independence of Xi and Xt, t 6= i.
28
![Page 29: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/29.jpg)
Immediate consequence:
E|W −W ∗| ≤ 1
σ2
n∑i=1
σ2i {E|Xi| + E|X∗
i |}
and
Proposition 1 Let X1, . . . , Xn be independent mean
zero variables with variances σ21, . . . , σ
2n and finite third
moments, and let W = (X1 + . . . + Xn)/σ where σ2 =
σ21 + . . . + σ2
n. Then for all absolutely continuous test
functions h,
|Eh(W )−Nh| ≤ 2||h′||σ3
n∑i=1
E|Xi| +
1
2|Xi|3
σ2i ,
29
![Page 30: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/30.jpg)
so in particular, when the variables are identically dis-
tributed with variance 1,
|Eh(W )−Nh| ≤ 3||h′||E|X1|3√n
.
30
![Page 31: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/31.jpg)
General construction: (Goldstein + R. 1997, Gold-
stein 2003)
Let Y ′, Y ′′ be exchangeable pair with distribution F (y′, y′′)
such that
E(Y ′′|F) = (1− λ)Y ′
for someF such that σ(Y ′) ⊂ F , and for some 0 < λ < 1
Let Y ′, Y ′′ have distribution
dG(y′, y′′) =(y′ − y′′)2
E(Y ′ − Y ′′)2dF (y′, y′′)
31
![Page 32: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/32.jpg)
and let U ∼ U(0, 1) be independent of Y ′, Y ′′, then
Y ∗ = UY ′ + (1− U)Y ′′
has the Y ∗-distribution.
If in addition Y ′ = V + T ′ and Y ′′ = V + T ′′ for
some T ′, T ′′, and on the same state space Y ′ = V + T ′
and Y ′′ = V + T ′′ for some T ′, T ′′ with |T ′| ≤ B and
|T ′′| ≤ B, then we can couple such that
|Y ′ − Y ∗| ≤ 3B.
32
![Page 33: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/33.jpg)
Example: Simple random sampling
Population of characteristics A with ∑a∈A a = 0 and
|a| ≤ n−1/2B for all a ∈ A. Let
X ′, X ′′, X2, . . . , Xn
be a simple random sample of size n + 1
Use notation ‖ Z ‖= ∑z∈Z z, put
Y ′ =‖ X′ ‖, Y ′′ =‖ X′′ ‖
So Y ′ − Y ′′ = X ′ −X ′′
Choose X ′, X ′′ ∝ (x′ − x′′)2I({x′, x′′} ∈ A33
![Page 34: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/34.jpg)
Intersection R = {X ′, X ′′} ∩ {X2, . . . , Xn} and
V =‖ {X2, . . . , Xn} \ R ‖; T ′ =‖ X ′ ∩R ‖
Let S be a simple random sample of size |R| from A \
{X ′, X ′′, X2, . . . , Xn} and put T ′ =‖ X ′∩S ‖; similarly
for T ′′, T ′′
As |R| ≤ 2 we have |T ′| ≤ 2n−1/2B and |T ′′| ≤ 2n−1/2B
When third moments vanish, fourth moments exist:
Order n−1 bound
34
![Page 35: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/35.jpg)
Also: Berry-Esseen bound, combinatorial central limit
theorem (Goldstein 2004)
35
![Page 36: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/36.jpg)
Lecture 2: Other distributions
Recap
Would like bounds on distributional distance
Use test functions to assess distance
36
![Page 37: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/37.jpg)
For standard normal distribution N (0, 1), distribu-
tion function Φ(x) Stein (1972, 1986)
1. Z ∼ N (0, 1) if and only if for all smooth functions f
Ef ′(Z) = EZf (Z)
2. For any smooth function h there is a smooth function
f = fh solving the Stein equation
h(x)−∫hdΦ = f ′(x)− xf (x)
(and bounds on f in terms of h)
37
![Page 38: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/38.jpg)
3. For any random variable W , smooth h
Eh(W )−∫hdΦ = Ef ′(W )− EWf (W )
Use Taylor expansion or couplings to quantify weak
dependence
If W ≥ 0,EW > 0 then W s has the W -size biased
distribution if
EWf (W ) = EWEf (W s)
for all f for which both sides exist
38
![Page 39: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/39.jpg)
Construction(Goldstein + Rinott 1996)
Suppose W = ∑ni=1 Xi with Xi ≥ 0, EXi > 0, all i.
Choose index V ∝ EXv. If V = v: replace Xv by Xsv
having the Xv-size biased distribution, independent, and
if Xsv = x: adjust Xu, u 6= v, such that
L(Xu, u 6= v) = L(Xu, u 6= v|Xv = x)
Then W s = ∑u 6=V Xu + Xs
V
Example: Xi ∼ Be(pi) for i = 1, . . . , n
Then W s = ∑u 6=V Xu + 1
39
![Page 40: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/40.jpg)
3. General situation
Target distribution µ
1. Find characterization: operatorA such that X ∼ µ if
and only if for all smooth functions f , EAf (X) = 0
2. For each smooth function h find solution f = fh of
the Stein equation
h(x)−∫hdµ = Af (x)
40
![Page 41: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/41.jpg)
3. Then for any variable W ,
Eh(W )−∫hdµ = EAf (W )
Usually need to bound f, f ′, or ∆f
Here: h smooth test function; for nonsmooth functions:
see techniques used by Shao, Chen, Rinott and Rotar,
Gotze
41
![Page 42: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/42.jpg)
The generator approach
Barbour 1989, 1990; Gotze 1993
Choose A as generator of a Markov process with sta-
tionary distribution µ, that is:
Let (Xt)t≥0 be a homogeneous Markov process
Put Ttf (x) = E(f (Xt)|X(0) = x)
Generator Af (x) = limt↓01t (Ttf (x)− f (x))
42
![Page 43: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/43.jpg)
Facts (see Ethier and Kurtz (1986), for example)
1. µ stationary distribution then X ∼ µ if and only if
EAf (X) = 0 for f for which Af is defined
2. Tth− h = A (∫ t0 Tuhdu) and formally
∫hdµ− h = A
(∫ ∞0 Tuhdu
)
if the r.h.s. exists
43
![Page 44: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/44.jpg)
Examples
1. Ah(x) = h′′(x)−xh′(x) generator of Ornstein-Uhlenbeck
process, stationary distribution N (0, 1)
2. Ah(x) = λ(h(x + 1)−h(x)) + x(h(x− 1)−h(x)) or
Af (x) = λf (x + 1)− xf (x)
Immigration-death process, immigration rate λ, unit
per capita death rate; stationary distribution Poisson(λ)
Advantage: generalisations to multivariate, diffusions, mea-
sure space...
44
![Page 45: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/45.jpg)
4. Chisquare distributions
Generator for χ2p:
Af (x) = xf ′′(x) +1
2(p− x)f ′(x)
(Luk 1994: Gamma(r, λ) ) A is the generator of a
Markov process given by the solution of the stochastic
differential equation
Xt = x +1
2
∫ t0 (p−Xs)ds +
∫ t0
√2XsdBs
where Bs is standard Brownian motion
45
![Page 46: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/46.jpg)
Stein equation
(χ2p) h(x)− χ2
ph = xf ′′(x) +1
2(p− x)f ′(x)
where χ2ph is the expectation of h under the χ2
p-distribution
46
![Page 47: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/47.jpg)
Lemma 1 (Pickett 2002)
Suppose h : R → R is absolutely bounded, |h(x)| ≤
ceax for some c > 0 a ∈ R, and the first k deriva-
tives of h are bounded. Then the equation (χ2p) has a
solution f = fh such that
‖ f (j) ‖≤√
2π√
p‖ h(j−1) ‖
with h(0) = h.
(Improvement over Luk 1994 in 1√p)
47
![Page 48: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/48.jpg)
Example: squared sum (R. + Pickett)
Xi, i = 1, . . . , n i.i.d. mean zero, variance one, exisiting
8th moment
S =1√n
n∑i=1
Xi
and
W = S2
Want
2EWf ′′(W ) + E(1−W )f ′(W )
48
![Page 49: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/49.jpg)
Put
g(s) = sf ′(s2)
then
g′(s) = f ′(s2) + 2s2f ′′(s2)
and
2EWf ′′(W ) + E(1−W )f ′(W )
= Eg′(S)− Ef ′(W ) + E(1−W )f ′(W )
= Eg′(S)− ESg(S)
Now proceed as in N (0, 1):
49
![Page 50: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/50.jpg)
Put
Si =1√n
∑j 6=i
Xj
Then by Taylor expansion, some 0 < θ < 1,
ESg(S) =1√n
n∑i=1
EXig(S)
=1√n
n∑i=1
EXig(Si) +1
n
n∑i=1
EX2i g
′(Si) + R1
where
R1 =1
n3/2
∑iEX3
i g′′(Si)
+1
2n2
∑iEX4
i g(3)
Si + θXi√n
50
![Page 51: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/51.jpg)
From independence
ESg(S) =1
n
n∑i=1
Eg′(Si) + R1
= Eg′(S) + R1 + R2
where
R2 =1
n3/2
∑iEXig
′′(Si)
+1
2n2
∑iEX2
i g(3)
Si + θXi√n
=1
2n2
∑iEX2
i g(3)
Si + θXi√n
by Taylor expansion, some 0 < θ < 1
51
![Page 52: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/52.jpg)
Bounds on R1, R2
Calculate
g′′(s) = 6sf ′′(s2) + 4s3f (3)(s2)
and
g(3)(s) = 24s2f (3)(s2) + 6f ′′(s2) + 8s4f (4)(s2)
so with βi = EX i1
52
![Page 53: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/53.jpg)
1
2n2
∑iEX2
i
∣∣∣∣∣∣∣∣g(3)(Si + θ
Xi√n
)
∣∣∣∣∣∣∣∣
≤ 24
n‖ f (3) ‖
1 +β4
n
+6
n‖ f ′′ ‖
+8
n‖ f (4) ‖
6 +β4
n+ 4
β23√n
+ 6β4
n+
β6
n2
= c(f )1
n.
Similarly for 12n2
∑i EX4
i
∣∣∣∣∣g(3)(Si + θ Xi√n)
∣∣∣∣∣, employ β8
53
![Page 54: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/54.jpg)
For 1n3/2
∑i EX3
i g′′(Si) have, for some c(f )
1
n3/2
∑iEX3
i g′′(Si) =
1√nβ3Eg′′(S) + c(f )
1
n
and
Eg′′(S) = 6ESf ′′(S2) + 4ES3f (3)(S2)
Note that g′′ is antisymmetric, g′′(−s) = −g′′(s), so
for Z ∼ N (0, 1) we have
Eg′′(Z) = 0
(Almost) routine now to show that |Eg′′(S)| ≤ c(f )/√
n
for some c(f ).
54
![Page 55: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/55.jpg)
Combining these bounds show: the bound on the dis-
tance to Chisquare(1) for smooth test functions is of
order 1n
Also: Pearson’s chisquare statistic
55
![Page 56: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/56.jpg)
5. Discrete Gibbs measure (R. + Eichelsbacher)
Let µ be a probability measure with support supp(µ) =
{0, . . . , N}, where N ∈ N0 ∪ {∞}. Write as
µ(k) =1
Zexp(V (k))
ωk
k!, k = 0, 1, . . . , N,
with Z = ∑Nk=0 exp(V (k))ωk
k! , where ω > 0 is fixed
Assume Z exists
Example: Po(λ)
ω = λ, V (k) = −λ, k ≥ 0, Z = 1
or V (k) = 0, ω = λ, Z = eλ
56
![Page 57: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/57.jpg)
For a given probability distribution (µ(k))k∈N0
V (k) = log µ(k) + log k! + log Z− k log ω, k = 0, 1, . . . , N,
with V (0) = log µ(0) + log Z
To each such Gibbs measure associate a birth-death
process:
57
![Page 58: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/58.jpg)
unit per-capita death rate dk = k
birth rate
bk = ω exp{V (k + 1)− V (k)} = (k + 1)µ(k + 1)
µ(k),
for k, k + 1 ∈ supp(µ)
then invariant measure µ
generator
(Ah)(k) = (h(k + 1)− h(k)) exp{V (k + 1)− V (k)}ω
+k(h(k − 1)− h(k))
58
![Page 59: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/59.jpg)
or
(Af )(k) = f (k + 1) exp{V (k + 1)− V (k)}ω − kf (k)
59
![Page 60: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/60.jpg)
Examples
1. Poisson-distribution with parameter λ > 0: We use
ω = λ, V (k) = −λ,Z = 1. The Stein-operator is
(Af )(k) = f (k + 1) λ− kf (k)
2. Binomial-distribution with parameters n and 0 <
p < 1: We use ω = p1−p, V (k) = − log((n− k)!), and
Z = (n!(1− p)n)−1. The Stein-operator is
(Af )(k) = f (k + 1)p(n− k)
(1− p)− kf(k).
60
![Page 61: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/61.jpg)
Bounds
Solution of Stein equation f for h: f (0) = 0, f (k) = 0
for k 6∈ supp(µ), and
f (j + 1) =j!
ωj+1e−V (j+1)
j∑k=0
eV (k)ωk
k!
(h(k)− µ(h)) .
Lemma 2 1. Put
M := sup0≤k≤N−1
max(eV (k)−V (k+1),
eV (k+1)−V (k)).
61
![Page 62: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/62.jpg)
Assume M < ∞. Then for every j ∈ N0:
|f (j)| ≤ 2 min
1,√
M√ω
.
2. Assume that the birth rates are non-increasing:
eV (k+1)−V (k) ≤ eV (k)−V (k−1),
and death rates are unit per capita. For every j ∈
N0
|∆f (j)| ≤ 1
j∧ eV (j)
ωeV (j+1).
62
![Page 63: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/63.jpg)
Example: Poisson-distribution with parameter λ >
0: non-uniform bound
|∆f (k)| ≤ 1
k∧ 1
λ,
leads to 1 ∧ 1/λ, see Barbour, Holst, Janson 1992
‖ f ‖≤ 2 min(1, 1√
λ
).
as in Barbour, Holst, Janson 1992
63
![Page 64: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/64.jpg)
Size-Bias coupling
Recall: W ≥ 0, EW > 0 then W ∗ has the W -size
biased distribution if
EWg(W ) = EWEg(W ∗)
for all g for which both sides exist, so
E{e(V (k+1)−V (k) ω g(X + 1)−X g(X)}
= E{e(V (k+1)−V (k) ω g(X + 1)− EXEg(X∗)}
and
EX = ωEeV (X+1)−V (X)
64
![Page 65: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/65.jpg)
Lemma 3 Let X ≥ 0 be such that 0 < E(X) < ∞,
let µ be a discrete Gibbs measure. Then X ∼ µ if and
only if for all bounded g
ω EeV (X+1)−V (X)g(X + 1)
= ω EeV (X+1)−V (X)Eg(X∗).
For any W ≥ 0 with 0 < EW < ∞
Eh(W )− µ(h)
= ω{EeV (W+1)−V (W )g(W + 1)
−EeV (W+1)−V (W )Eg(W ∗)}65
![Page 66: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/66.jpg)
where g is the solution of the Stein equation.
66
![Page 67: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/67.jpg)
Can also compare two discrete Gibbs distributions by
comparing their birth rates and their death rates (see also
Holmes)
Example: Poisson(λ1) and Poisson(λ2) gives
|Eh(X)−∫hdµ| ≤ ‖ f ‖ |λ− λ2|
67
![Page 68: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/68.jpg)
6. Final remarks
If X1, X2, . . . , Xn i.i.d., P(Xi = 1) = p = 1−P(Xi =
0), using Stein’s method we can show that
supx|P ((np(1− p))−1/2 n∑
i=1(Xi − p) ≤ x)
−P (N (0, 1) ≤ x)|
≤ 6
√√√√√√p(1− p)
n
and
supx|P (
n∑i=1
Xi = x)− P (Po(np) = x)|
≤ min(np2, p)
68
![Page 69: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/69.jpg)
So, if p < 36n+36, the bound on the Poisson approximation
is smaller than the bound on the normal approximation
69
![Page 70: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/70.jpg)
• Exchangeable pair couplings, also used for variance
reduction in simulations
• Multivariate, also coupling approaches
• General distributional transformations
• Bounds in the presence of dependence
• In the i.i.d. case: Berry-Esseen inequality not quite
recovered
70
![Page 71: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/71.jpg)
Further reading
1. Arratia, R., Goldstein, L., and Gordon, L. (1989). Two moments suffice for Poisson approx-
imation: The Chen-Stein method. Ann. Probab. 17, 9-25.
2. Barbour, A.D. (1990). Stein’s method for diffusion approximations, Probability Theory and
Related Fields 84 297-322.
3. Barbour, A.D., Holst, L., and Janson, S. (1992). Poisson Approximation. Oxford University
Press.
4. Barbour, A.D. and Chen. L.H.Y. eds (2005). An Introduction to Stein’s Method. Lecture
Notes Series 4, Inst. Math. Sciences, World Scientific Press, Singapore.
5. Barbour, A.D. and Chen. L.H.Y. eds (2005). Stein’s Method and Applications. Lecture
Notes Series 5, Inst. Math. Sciences, World Scientific Press, Singapore.
6. Diaconis, P., and Holmes, S. (eds.) (2004). Stein’s Method: Expository Lectures and Appli-
cations. IMS Lecture Notes 46.
7. Raic, M. (2003). Normal approximations by Stein’s method. In Proceedings of the Seventh
Young Statisticians Meeting, Mvrar, A. (ed.), Metodoloski zveski, 21, Ljubljana, FDV, 71-97.
71
![Page 72: Gesine Reinert Department of Statistics University of Oxfordreinert/talks/steinintroduction.pdf · 2008-01-16 · A Short Introduction to Stein’s Method Gesine Reinert Department](https://reader036.fdocuments.in/reader036/viewer/2022070900/5f3f64a1fc08093e7c56f6fb/html5/thumbnails/72.jpg)
8. Stein, C. (1972). A bound for the error in the normal approximation to the distribution of a
sum of dependent random variables. Proc. Sixth Berkeley Symp. Math. Statist. Probab. 2
583-602, Univ. California Press, Berkeley.
9. Stein, C. (1986). Approximate Computation of Expectations. IMS, Hayward, CA.
72