0% found this document useful (0 votes)
13 views6 pages8 107
The document discusses statistical concepts including sampling distributions, maximum likelihood estimators, and the central limit theorem. It provides exercises related to binomial and Poisson distributions, as well as asymptotic distributions of estimators. Historical contributions to statistical theory by notable figures such as R. A. Fisher and J. Neyman are also mentioned.
Uploaded by
taskmask293Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
13 views6 pages8 107
The document discusses statistical concepts including sampling distributions, maximum likelihood estimators, and the central limit theorem. It provides exercises related to binomial and Poisson distributions, as well as asymptotic distributions of estimators. Historical contributions to statistical theory by notable figures such as R. A. Fisher and J. Neyman are also mentioned.
Uploaded by
taskmask293Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 6
senor iv. s
N(O.)
7 0
(2) Sampling from 72.2)
SSL. BEC SUGOTE Sr SM MKERGR ESneators
—
t = 0 1
(b) Sampling irom U-shaped distribution
FIGURE 10.6-2: Convergence of moment-generating functions
EXERCISES
(06-1.
106
06-3,
10.7
Let ¥ be the number of defectives in a box of 501
articles taken from the outpet of a machine. Eack
article is detestive with probability 0.01. Find the
probability that ¥ = 0,1,2, or 3
(a) By using the binomial distribution.
(b) By using the Poisson appreximation.
. The probability that a certain type of inoculation
takes effect is 0.995. Use the Poisson distribution
to approximate the probability that at most 2 out
‘01 400 people given the inoculation find that it
has not taken effect.
Hivt: Let p = | ~ 0.995 = 0.003,
Let S? be the sample variance ofa randont sample
of size from N(u. 7). Show thar the limit, as
106-4,
1065.
11 00, 01 the moment-generating funetion of 5?
is °° Thus, m the limit, the distribution of S? is
degenerate with probability 1 at a2
Let ¥ be 4212). Use the central limit theo-
rem to demonstrate that Wo= (¥ — nj/V2i
has @ limiting distribution funetion that is
N(0.1).
Hiny. Think of ¥ as being the sum of a sandom
sample from a certain distributiou.
Let ¥ have a Poisson distribution ‘with mean
3n. Use the central limit theorem %9 show that
W = (¥ ~ 3n)/V3n bes a limiting distribution
thanis M(0,1)
ASYMPTOTIC DISTRIBUTIONS OF MAXIMUM LIKELIHOOD ESTIMATORS
Let us consider e distribution cf the continuous type with p.d.f. f(x: 0) such that the
parameter @ is not involved in the support of the distribution. Moreover, we want
fx: 8) to possess a number of mathematical properties that we do not list here
However, in particular, we want to be able to find the maximum likelihood estimator
@ by solving
OI <9,
7
where here we wse a partial-derivative sign because L(8) involves ry, 42,....%q. 100,
That is,
aiiaLO) _ 9
36
where naw, with 0 in this expression, LIB) = f(X1: O)F(X>: BY f(Xus 8),
‘We can approximate the left-hand member cf this fatter equation by a linear function found from the fst two terms of a Taylor's series expanded about 4
namely,
AML). 3
0
when L(6) = f( Xi: O)f(X2: OF f( Xn: 8). .
Obviously. this approximation is good enough only if @ is close 40 0, and an
adequate mathematical proof involves those conditions, which we have not given
here. (See Hogg, McKean, and Craig, 2005.) But a heuristic argument can be made
by solving for — 00 obtain
alin L(#)]
a-6 ao \
_& [In La) (ort)
0"
Recall that
In L(0) = Inf(X:8) + InfiWe) +o + Inf Need
and
KO) _ & afinf(X0)]
ape 10.7.2
ow 72)
which is the numerator in Equation 10.7-1, However, Equation 10.7-2
of the » independent and identically distributed random Variables
ol i= 1,2,
ves the sum,
alln (Xi
¥,
and thus, by the central limit theocem, has an approximate normal distribution sith
‘mean (in the continuous case) equal to-
I AAEM ex 9) ae -f- alf
a0
a0
-f AOI 4,
~ 6
oy [fn 8) de
‘Clearly, we need a certain mathematical condition that makes it permissitle to
interchange the operations of in(egration and differentiation in those fast steps, Of
course, the integral of f(t: 0) is equal to Lbecause itis a pd
Since we now know that the mean of each Y is
r afin flx:@)]
4 f(a.d) dx = 0,
cn) Section 10.7 Asymptotic Distributions of Maximum Liketihood Estimators 537
let ws take derivatives of each member of this equation with respect to 8. obtaining
ifin f(x: @} abe: 8}
However,
Abas) AllnfeOL gg),
0 ao
r {nde ON 5 oar = fr
(x. 0) dx
Since: £(¥) = 0. this last expression provides the variance of Y = afin f(.X:6)]/00.
‘Then the variance of the sum in Equation 16.7-2 is n times this value, namely
-ne {Plat x a
ae
Let us rewrite Equation 10.7-1 as
alin (08
nb (alin fi X.0) Vee}
TlInL))
a
Fc#lin FX: 0) 708)
£ (197-3)
(Ef
The numerator of the right-hand member of Equation 10.7-3 has an approximate
N(G,U) distribution, and the aforementioned unstated mathematical conditions,
require, in some sense, that
[In L(0 Pl ;
1 afin t40)) E{ fin F(X: 0)]/ 002}
nw
Accordingly. the ratios given in Equation 10.7-3 must be approximately N(O. |). That
is, 0 has an approximate normal distribution with mean @ and standard deviation
1
Jan E (Phin (CX: 8) 08? |
SENEERORAT (continuation of Exanple 6.1-1), With the underlying exponential peli
fasd)= Fev 0 << oy GE N= (#0 <0 <0)
6
X is tite maximum likeliliood estimator, Since:
Inf(x;e) = — Ine
0
and
aoe Ol 1 ye gg ere) 2x
i oe 0 eo ——
we have .
ox 2
le” ee
because F(X) = 0. That is. X has an approximate normal distribution with mean
@ and standard deviation 6//n. Thus. the random interval ¥ + 1.96(6// ) has an
approximate probability of 0.95 that it covers @. Substituting the observed X for @. as
well as for X. we say that ¥ £ 1,96X//n is an approximate 05% confidence interval
lor 0. .
While the development of the preceding result used a continuous-type distribution,
the result holds for the diserete type also. as long as the support does not involve the
parameter. This is illustrated in the next example.
Set
(continuation of Exercise 6.1-3). I the random sample arises from a Poisson distribu-
tion with pet.
aten*
fea) ==$-, r= 0,1,2, Aeon
{A:0
