You are on page 1of 19

1

12. Principles of Parameter Estimation


The purpose of this lecture is to illustrate the usefulness of
the various concepts introduced and studied in earlier
lectures to practical problems of interest. In this context,
consider the problem of estimating an unknown parameter
of interest from a few of its noisy observations. For
example, determining the daily temperature in a city, or the
depth of a river at a particular spot, are problems that fall
into this category.
Observations (measurement) are made on data that contain
the desired nonrandom parameter u and undesired noise.
Thus, for example,
PILLAI
2
(12-1) noise, part) (desired signal n Observatio + =
or, the i th observation can be represented as

Here u represents the unknown nonrandom desired
parameter, and represent random variables
that may be dependent or independent from observation to
observation. Given n observations
the estimation problem is to obtain the best estimator for
the unknown parameter u in terms of these observations.
Let us denote by the estimator for u . Obviously is
a function of only the observations. Best estimator in what
sense? Various optimization strategies can be used to define
the term best.
. , , 2 , 1 , n i n X
i i
= + =u
(12-2)
, , , ,
2 2 1 1 n n
x X x X x X = = =
) (

X u ) (

X u
n i n
i
, , 2 , 1 , =
PILLAI
3
Ideal solution would be when the estimate coincides
with the unknown u . This of course may not be possible,
and almost always any estimate will result in an error given
by

One strategy would be to select the estimator so as to
minimize some function of this error - such as -
minimization of the mean square error (MMSE), or
minimization of the absolute value of the error etc.
A more fundamental approach is that of the principle of
Maximum Likelihood (ML).
The underlying assumption in any estimation problem is
(12-3)
. ) (

u u = X e
) (

X u
) (

X u
PILLAI
4
that the available data has something to do
with the unknown parameter u . More precisely, we assume
that the joint p.d.f of given by
depends on u. The method of maximum likelihood assumes
that the given sample data set is representative of the
population and chooses that value for u
that most likely caused the observed data to occur, i.e., once
observations are given, is a
function of u alone, and the value of u that maximizes the
above p.d.f is the most likely value for u , and it is chosen as
the ML estimate for u (Fig. 12.1).
n
X X X , , ,
2 1

n
X X X , , ,
2 1
) ; , , , (
2 1
u
n X
x x x f
), ; , , , (
2 1
u
n X
x x x f
n
x x x , , ,
2 1
) ; , , , (
2 1
u
n X
x x x f
) (

X
ML
u
) (

X
ML
u
) ; , , , (
2 1
u
n X
x x x f
u
Fig. 12.1
PILLAI
5
Given the joint p.d.f
represents the likelihood function, and the ML estimate can
be determined either from the likelihood equation

or using the log-likelihood function (sup in (12-4)
represents the supremum operation)

If is differentiable and a supremum
exists in (12-5), then that must satisfy the equation

We will illustrate the above procedure through several
examples:
, , , ,
2 2 1 1 n n
x X x X x X = = =
) ; , , , (
2 1
u
n X
x x x f
) ; , , , ( sup
2 1

u
u
n X
x x x f
ML

(12-4)
). ; , , , ( log ) ; , , , (
2 1 2 1
u u
n X n
x x x f x x x L = (12-5)
) ; , , , (
2 1
u
n
x x x L
A
ML
u

. 0
) ; , , , ( log

2 1
=
c
c
=
ML
n X
x x x f
u u
u
u
(12-6)
PILLAI
6
Example 12.1: Let represent n
observations where u is the unknown parameter of interest,
and are zero mean independent normal r.vs with
common variance Determine the ML estimate for u.
Solution: Since are independent r.vs and u is an unknown
constant, we have s are independent normal random
variables. Thus the likelihood function takes the form

Moreover, each is Gaussian with mean u and variance
(Why?). Thus

Substituting (12-8) into (12-7) we get the likelihood function
to be
, 1 , n i w X
i i
= + =u
.
2
o
, 1 , n i w
i
=
i
X
i
w
. ) ; ( ) ; , , , (
1
2 1 [
=
=
n
i
i X n X
x f x x x f
i
u u
(12-7)
.
2
1
) ; (
2 2
2 / ) (
2
o u
to
u

=
i
i
x
i X
e x f
i
X
2
o
(12-8)
PILLAI
7
.
) 2 (
1
) ; , , , (
1
2 2
2 / ) (
2 / 2
2 1

=
=

n
i
i
x
n
n X
e x x x f
o u
to
u
(12-9)
It is easier to work with the log-likelihood function
in this case. From (12-9)

and taking derivative with respect to u as in (12-6), we get

or

Thus (12-12) represents the ML estimate for u , which
happens to be a linear estimator (linear function of the data)
in this case.
, 0
2
) (
2
) ; , , , ( ln
1
2

2 1
=

=
c
c
=
=
=

ML
ML
n
i
i n X
x x x x f
u u
u u
o
u
u
u
(12-10)
.
1
) (

=
=
n
i
i ML
X
n
X u
(12-11)
(12-12)
,
2
) (
) 2 ln(
2
) ; , , , ( ln ) ; (
1
2
2
2
2 1
=

= =
n
i
i
n X
x n
x x x f X L
o
u
to u u
) ; ( u X L
PILLAI
8
Notice that the estimator is a r.v. Taking its expected value,
we get

i.e., the expected value of the estimator does not differ from
the desired parameter, and hence there is no bias between
the two. Such estimators are known as unbiased estimators.
Thus (12-12) represents an unbiased estimator for u .
Moreover the variance of the estimator is given by



The later terms are zeros since and are independent
r.vs.
, ) (
1
)] (

[
1
u u = =

=
n
i
i ML
X E
n
x E
(12-13)
. ) )( ( ) (
1

1
] )

[( )

(
1 , 1 1
2
2
2
1
2
2
)
`

+ =

|
.
|

\
|
= =

= = = =
=
n
i
n
j i j
j i
n
i
i
n
i
i ML ML
X X E X E
n
X E
n
E Var
u u u
u u u u
i
X
j
X
PILLAI
9
Then

Thus

another desired property. We say such estimators (that
satisfy (12-15)) are consistent estimators.
Next two examples show that ML estimator can be highly
nonlinear.
Example 12.2: Let be independent, identically
distributed uniform random variables in the interval
with common p.d.f
(12-14)
. ) (
1
)

(
2
2
2
1
2
n n
n
X Var
n
Var
n
i
i ML
o o
u = = =

=
, as 0 )

( n Var
ML
u
n
X X X , , ,
2 1

) , 0 ( u
, 0 ,
1
) ; ( u
u
u < < =
i i X
x x f
i
PILLAI
(12-15)
(12-16)
10
where u is an unknown parameter. Find the ML estimate
for u.
Solution: The likelihood function in this case is given by


From (12-17), the likelihood function in this case is
maximized by the minimum value of u , and since
we get

to be the ML estimate for u . Notice that (18) represents a
nonlinear function of the observations. To determine
whether (12-18) represents an unbiased estimate for u , we
need to evaluate its mean. To accomplish that in this case, it
is easier to determine its p.d.f and proceed directly. Let
. ) , , , max( 0 ,
1

1 , 0 ,
1
) ; , , , (
2 1
2 2 1 1
u
u
u
u
u
s s =
= s < = = = =
n
n
i
n
n n X
x x x
n i x x X x X x X f

(12-17)
), , , , max(
2 1 n
X X X > u
) , , , max( ) (

2 1 n ML
X X X X = u (12-18)
PILLAI
11
(12-19)
) , , , max(
2 1 n
X X X Z =
with as in (12-16). Then


so that

Using (12-21), we get

In this case and hence the ML estimator is not
an unbiased estimator for u . However, from (12-22) as
i
X
, 0 , ) ( ) (
) , , , ( ] ) , , , [max( ) (
1 1
2 1 2 1
u
u
< <
|
.
|

\
|
= = s =
s s s = s =
[ [
= =
z
z
z F z X P
z X z X z X P z X X X P z F
n
n
i
X
n
i
i
n n Z
i

(12-20)

. otherwise , 0
, 0 ,
) (
1

< <
=

u
u
z
nz
z f
n
n
Z
(12-21)
.
) / 1 1 ( 1
) ( ) ( )] (

[
1
0

0
n n
n
dz z
n
dz z f z Z E X E
n
n
n
n
Z ML
+
=
+
= = = =
+
} }
u
u
u
u
u
u u
(12-22)
, )] (

[ u u = X E
ML
n
PILLAI
12
i.e., the ML estimator is an asymptotically unbiased
estimator. From (12-21), we also get

so that

Once again as implying that the
estimator in (12-18) is a consistent estimator.
Example 12.3: Let be i.i.d Gamma random
variables with unknown parameters o and | . Determine the
ML estimator for o and | .
(12-23)
,
) / 1 1 (
lim )] (

[ lim u
u
u =
+
=

n
X E
n
ML
n
2
) ( ) (
2
0
1
0
2 2
+
= = =
} }
+
n
n
dz z
n
dz z f z Z E
n
n
Z
u
u
u u
(12-24)
.
) 2 ( ) 1 ( ) 1 ( 2
)] ( [ ) ( )] (

[
2
2
2
2 2 2
2 2
+ +
=
+

+
= =
n n
n
n
n
n
n
Z E Z E X Var
ML
u u u
u
0 )] (

[ X Var
ML
u , n
n
X X X , , ,
2 1

PILLAI
(12-25)
13
Solution: Here and

This gives the log-likelihood function to be


Differentiating L with respect to o and | we get



Thus from (12-29)
, 0 >
i
x
.
)) ( (
) , ; , , , (
1
1
2 1
1
[
=


I
=
=
n
i
x
i
n
n
n X
n
i
i
e x x x x f
|
o
o
o
|
| o
(12-26)
. log ) 1 ( ) ( log log
) , ; , , , ( log ) , ; , , , (
1 1
2 1 2 1

= =
|
.
|

\
|
+ I =
=
n
i
i
n
i
i
n X n
x x n n
x x x f x x x L
| o o | o
| o | o
(12-27)
, 0 log ) (
) (
log

, ,
1
= + I'
I
=
c
c
=
=

| o | o
o
o
|
o
n
i
i
x
n
n
L
(12-28)
. 0

, ,
1
= =
c
c
=
=

| o | o
|
o
|
n
i
i
x
n L
(12-29)
,
1

) (

=
=
n
i
i
ML
ML
x
n
X
o
|
(12-30)
PILLAI
14
and substituting (12-30) into (12-28), it gives

Notice that (12-31) is highly nonlinear in
In general the (log)-likelihood function can have more than
one solution, or no solutions at all. Further, the (log)-
likelihood function may not be even differentiable, or it can
be extremely complicated to solve explicitly
(see example 12.3, equation (12-31)).
Best Unbiased Estimator:
Referring back to example 12.1, we have seen that (12-12)
represents an unbiased estimator for u with variance given
by (12-14). It is possible that, for a given n, there may be
other
.
1 1
log
) (
) (
log
1 1

= =
|
.
|

\
|
=
I
I'

n
i
i
n
i
i
ML
ML
ML
x
n
x
n o
o
o
.
ML
o
(12-31)
PILLAI
15
unbiased estimators to this problem with even lower
variances. If such is indeed the case, those estimators will be
naturally preferable compared to (12-12). In a given scenario,
is it possible to determine the lowest possible value for the
variance of any unbiased estimator? Fortunately, a theorem by
Cramer and Rao (Rao 1945; Cramer 1948) gives a complete
answer to this problem.
Cramer - Rao Bound: Variance of any unbiased estimator
based on observations for u must
satisfy the lower bound

This important result states that the right side of (12-32) acts
as a lower bound on the variance of all unbiased estimator for
u, provided their joint p.d.f satisfies certain regularity
restrictions. (see (8-79)-(8-81), Text).
u

n n
x X x X x X = = = , , ,
2 2 1 1

.

) ; , , , ( ln
1

) ; , , , ( ln
1
)

(
2
2 1
2
2
2 1
|
|
.
|

\
|
c
c

=
|
.
|

\
|
c
c
>
u
u
u
u
u
n X
n X
x x x f
E
x x x f
E
Var

(12-32)
PILLAI
16
Naturally any unbiased estimator whose variance coincides
with that in (12-32), must be the best. There are no better
solutions! Such estimates are known as efficient estimators.
Let us examine whether (12-12) represents an efficient
estimator. Towards this using (12-11)


and




and substituting this into the first form on the right side of
(12-32), we obtain the Cramer - Rao lower bound for this
problem to be
; ) (
1

) ; , , , ( ln
2
1
4
2
2 1
|
.
|

\
|
=
|
.
|

\
|
c
c

=
n
i
i
n X
X
x x x f
u
o u
u
(12-33)
,
1

)] )( [( ] ) [(
1

) ; , , , ( ln
2
1
2
4
1 , 1 1
2
4
2
2 1
o
o
o
u u u
o u
u
n
X X E X E
x x x f
E
n
i
n
i
n
j i j
j i
n
i
i
n X
= =
)
`

+ =
|
.
|

\
|
c
c


=
= = = =

(12-34)
PILLAI
17
PILLAI


But from (12-14) the variance of the ML estimator in (12-12)
is the same as (12-35), implying that (12-12) indeed represents
an efficient estimator in this case, the best of all possibilities!

It is possible that in certain cases there are no unbiased
estimators that are efficient. In that case, the best estimator
will be an unbiased estimator with the lowest possible
variance.
How does one find such an unbiased estimator?
Fortunately Rao-Blackwell theorem (page 335-337, Text)
gives a complete answer to this problem.
Cramer-Rao bound can be extended to multiparameter case
as well (see page 343-345,Text).
.
2
n
o
(12-35)
18
So far, we discussed nonrandom parameters that are
unknown. What if the parameter of interest is a r.v with
a-priori p.d.f How does one obtain a good estimate
for u based on the observations

One technique is to use the observations to compute its
a-posteriori probability density function
Of course, we can use the Bayes theorem in (11.22) to
obtain this a-posteriori p.d.f. This gives



Notice that (12-36) is only a function of u , since
represent given observations. Once again, we can look for
? ) (u
u
f
? , , ,
2 2 1 1 n n
x X x X x X = = =
). , , , | (
2 1 | n X
x x x f u
u
.
) , , , (
) ( ) | , , , (
) , , , | (
2 1
2 1 |
2 1 |
n X
n X
n X
x x x f
f x x x f
x x x f

u u
u
u u
u
=
(12-36)
n
x x x , , ,
2 1

PILLAI
19
the most probable value of u suggested by the above
a-posteriori p.d.f. Naturally, the most likely value for u is
that corresponding to the maximum of the a-posteriori p.d.f
(see Fig. 12.2). This estimator - maximum of the a-posteriori
p.d.f is known as the MAP estimator for u .
It is possible to use other optimality criteria as well. Of
course, that should be the subject matter of another course!
PILLAI
MAP
u

) , , , | (
2 1 n
x x x f u
u
Fig. 12.2

You might also like