Professional Documents
Culture Documents
Exercise 1
We have
P ois(x|) = e
x
x!
e
()
P ois(x|)(; , )
(|x; , ) = P
i=0 P ois(x|)(; , )
(; , ) =
(1)
= e(+1) x+1
(2)
0 0 1 0
e
(0 )
0 = + x
(|x; 0 , 0 ) =
0 = + 1
Exercise 2
a) We first derive the expectation of the variance MLE. That is
"
#
N
2
1 X
E
MLE = E
(xi x
)2 =
N
i=1
"N
#
X
1
2
2
= E
xi + x
2
xxi
N
i=1
2
1
= (N E x2
N E x
) =
N
| {z }
| {z }
Var[x]+E[x]2
Var[
x]/N +E[
x ]2
1
N 1
(N Var[x] Var[x]) =
Var[x]
N
N
=
N 1
N 1 MLE
i=1
2
N
E
MLE =
N 1
= Var[x]
=
b)
P
i) As we have just shown, Bias( N 11 N
)2 , 2 ) = 0. We
i=1 (xi x
P
N
)2 ].
now turn our attention to Var[ N 11 i=1 (xi x
1
Var[(N 1)
N
X
(xi x
)2 ] = Var[(N 1)1 2 2N 1 ] =
i=1
= (N 1)2 4 Var[2N 1 ] =
= 2(N 1)1 4
N
X
N
X
(xi x
) ) = Var[c
(xi x
)2 ] + (c(N 1) 2 2 )2 =
2
i=1
i=1
1
N +1
We also have
MSE(c
PN
i=1 (xi
x
)2 )
Exercise 3
a) We assume that X Binomial(n, 0 ) From this, we obtain MSE(
p)
X
E [
p] = E
= 0
n
"
2 #
X
X
MSE(
p) = E
=
0
= Var
n
n
=
Var[X]
n0 (1 0 )
0 (1 0 )
=
=
n2
n2
n
h i
E = (0.2)(0.1) + (0.8)E [
p] = 0.02 + 0.80
h i
0 ) = E 0 = 0.2( 1 0 )
Bias(,
10
= Var[0.02 + 0.8
Var[]
p] = 0.64 Var[
p] = 0.64
0 (1 0 )
n
= Bias2 (,
0 ) + Var[]
=
MSE()
(1 0 )0
= (0.02 0.20 )2 + 0.64
n
We proceed with a comparison of the estimators.
MSE()
<1
MSE(
p)
>0
MSE(
p) MSE()
0 (1 0 )
(1 0 )0
(0.02 0.20 )2 + 0.64
>0
n
n
(n + 9)02 (0.2n + 9)0 + 0.01n < 0
This quadratic inequality in 0 indicates that when n = 25, outperforms
p if 0.018707 < 0 < 0.393058 and when n = 100, outperforms p if
0.0407128 < 0 < 0.225342
Exercise 4
a) The posterior pD|X (D|X) is
pD|X (D1 |X) =
0.06e 2 (x7)
1
D1
11
47000
X)) R
2
3
D0
D1
X R
D0
11 1
47000
+ ln(
)
2
3
3
(3)
(4)
(5)
Where in case of equality the choice between testing again and not can
be arbitrary for the expected cost is the same.
If the prior probability of having D is 105 we can plug in the different
prior in (3) to see that the first test should produce a value greater
than (or, at least, equal to) 3843.14 in order for it to be optimal
to perform a second test. Given the distribution of the first tests
statistic, even in presence of the disease, the probability of such an
event is extremely close to 0 and thus the situation is most unlikely to
arise in practice.