STSM 2626
BAYESIAN STATISTICAL
INFERENCE
2025
Notes prepared by Dr I. Garisch. Notes edited by Dr D. Chikobvu and Dr M. Sjölander
Slides (from notes) by Dr M. Sjölander
Presented by Dr M. Sjölander
=a
(We will apply all this new stuff in example 5.1.1.)
θ1 = “little defective items in lot” = { }
θ2 = “little non-def. items in lot” = { }
Loss function l(θk, di(xj)) = l(θk, am) ↗
Action space A State of nature Data i.e. X Examples of decision functions
a1 = sell lot θ1 = {p < p0} X1 = {� < c} � �� � = ��
�� � = �
a2 = junk lot θ2 = {p ≥ p0} X2 = {� ≥ c} �� �� � = ��
Examples of decision functions
� �� � = ��
�� � = �
�� �� � = ��
�� �� � = ��
�� � =
�� �� � = ��
�� �� � = ��
�� � =
�� �� � = ��
Concepts
• Action space A – the various actions we can take
• State of nature θ – state of the population / we cannot
see this
• Data – sample which we can see (and use to make
decisions on which action to take)
• Decision functions – for each possible value in the
sample/data we decide what action will be taken if the
value is observed
• Loss function – the losses we make for each specific
state of nature, and each specific action that we take
– Rules to pick best di (before a sample is observed i.e. prior analysis)
if θ = θ1: , then d1 is better.
e.g.
if θ = θ2: , then d2 is better.
i
i
i.e. we have a set: {max[R(θ,d1)],…, max[R(θ,dn)]}
θ∊ Θ θ∊ Θ
max[R(θ,d*)] = i
θ∊ Θ i
d*
^
)
θ
i.e. we have a set: {E[R(θ,d
θ 1)],…, E[R(θ,d
θ n)]}
E[R(θ,d**)] = min{E[R(θ,di)]}
θ di θ
because he guesses the stratum (hole) is 12m
Action space
because he guesses the stratum (hole) is 16m
i.e. the stratum (hole) is really 12m
State of nature
i.e. the stratum (hole) is really 16m
Loss function l(θk, di(xj)) = l(θk, am)
Data
X = xj
Machine says that the stratum (hole) is 12m =
Machine says that the stratum (hole) is 14m = Pdf of X
Machine says that the stratum (hole) is 16m =
di(xj) a1 = he guesses the stratum (hole) is 12m
a2 = he guesses the stratum (hole) is 16m
x1 = Machine says that the stratum (hole) is 12m
x2 = Machine says that the stratum (hole) is 14m
x3 = Machine says that the stratum (hole) is 16m
a) Find the minimax rule.
b) Find the Bayes rule if p(θ1) = 0.8 and p(θ2) = 0.2.
c) Find the Bayes rule if p(θ1) = ξ and p(θ2) = 1 – ξ.
Solution:
a)
(Remember: Expected Value is taken w.r.t. X)
We need to find R(θk , di) for all k and i.
di(xj)
l(θk, am)
}
R(θk,di) θ1 θ2 Max
d1 0 400 400
d2 40 40 40 Minimum is 40
d3 10 120 120
d4 100 0 100
b)
}
R(θk,di) θ1 θ2 B(di) = Eθ[R(θk,di)]
d1 0 400 (0)(0.8) + (400)(0.2) = 80
d2 40 40 (40)(0.8) + (40)(0.2) = 40
d3 10 120 (10)(0.8) + (120)(0.2) = 32 Minimum is 32
d4 100 0 (100)(0.8) +( 0)(0.2) = 80
c)
R(θk,di) θ1 θ2 B(di) = E[R(θk,di)]
d1 0 400 (0)(ξ) + (400)(1 – ξ) = 400 – 400ξ
Minimum depends
d2 40 40 (40)(ξ) + (40)(1 – ξ) = 40 on value of ξ, so
d3 10 120 (10)(ξ) + (120)(1 – ξ) = 120 – 110ξ we draw a graph.
d4 100 0 (100)(ξ) + (0)(1 – ξ) = 100ξ
The values of 0.4, 0.73
and 0.97 you get by
setting the various
B(di)’s to other B(dj)’s.
B(di) was the last
column in the table at
the bottom of the
previous slide. For
example, to get the
0.73 we would set B(d2)
= B(d3)
40 = 120 – 110ξ
110ξ = 80
ξ = 80/110 = 0.73
*
*
*
*