Section 7.7 (p.417)
1. E[ (1/n)SXik ] = (1/n) E[ SXik ] = (1/n) n E[ Xk
] = E[ Xk ]
2. (1/n) SXi2 is an unbiased estimator
of E(X2),
is an unbiased
estimator of Var(X). Therefore, (1/n) SXi2
+
is an unbiased
estimator of [E(X)]2.
4. Follow the hint, we have
![]()
Therefore,
![]()
Therefore, we have d(x) = 2X.
5. From equation (8) on p.415, the MSE of S02
is (2n-1)/n2 s4
and the MSE of S12 is 2/(n-1) s4.
Since (2n-1)/n2 < 2/(n-1) for every positive integer n, it
follows that the MSE of S02 is smaller than the MSE of S12
for all possible m and s2.
7. E[d(X)] = p Þ S d(x)
pqx = p Þ S d(x)
qx = 1
It follows that d(0) + d(1)q
+ d(2)q2 + … = 1 for all possible q.
Therefore the unbiased estimator d(X)
is d(0) = 1 and d(X)
= 0 for X > 0.
9. Let X be the number of failures that will occur before
exactly k successes have been obtained. From Section 5.5 we know that X has a
negative binomial distribution with parameters k and p. Also note that N = X +
k. Now we want to show E[(k-1)/(N-1)] = p.
![]()
The last summation is 1
because it is the sum of the probabilities of a negative binomial distribution
with parameters k-1 and p.
11. (a) Let X be the value of a certain
characteristic among the total population and Y be the stratum number. Then
m = E(X) = E[ E(X|Y)] = S E(X|Y=i ) Pr(Y=i ) = S mi pi
![]()
(b) ![]()
Note that n=Sni, we can let
. And set
. Solve theses equations we get

Note: The ni
obtained above are not necessarily integers. The answer should be the closest integer.
12. (a) E[d0(T)]
= E[E(d|T)] = E(d)
= q
(b) From Theorem 1. of Sec. 6.9, R(q, d0) £ R(q,
d).
R(q, d).=
E[(d-q)2]
= Varq(d)
because E(d)=q.
For the same reason, R(q, d0).= Varq(d0). Therefore Varq(d0) £ Varq(d).
13. Let F(y) be the cdf of Yn. Then
F(y) = Pr(Yn£y) = Pr(X1£y)… Pr(Xn£y) = (y/q)n.
Therefore Yn
has pdf f(y) =F¢(y) = nyn-1/qn.
