Chapter 16

Measurement Error Models

A fundamental assumption in all the statistical analysis is that all the observations are correctly measured. In
the context of multiple regression model, it is assumed that the observations on the study and explanatory
variables are observed without any error. In many situations, this basic assumption is violated. There can be
several reasons for such a violation.
 For example, the variables may not be measurable, e.g., taste, climatic conditions, intelligence,
education, ability etc. In such cases, the dummy variables are used, and the observations can be
recorded in terms of values of dummy variables.
 Sometimes the variables are clearly defined, but it is hard to take correct observations. For example,
the age is generally reported in complete years or in multiple of five.
 Sometimes the variable is conceptually well defined, but it is not possible to take a correct
observation on it. Instead, the observations are obtained on closely related proxy variables, e.g., the
level of education is measured by the number of years of schooling.
 Sometimes the variable is well understood, but it is qualitative in nature. For example, intelligence is
measured by intelligence quotient (IQ) scores.

In all such cases, the true value of the variable can not be recorded. Instead, it is observed with some error.
The difference between the observed and true values of the variable is called as measurement error or
errors-in-variables.

Difference between disturbances and measurement errors:

The disturbances in the linear regression model arise due to factors like the unpredictable element of
randomness, lack of deterministic relationship, measurement error in study variable etc. The disturbance
term is generally thought of as representing the influence of various explanatory variables that have not
actually been included in the relation. The measurement errors arise due to the use of an imperfect measure
of true variables.

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

1
Large and small measurement errors
If the magnitude of measurement errors is small, then they can be assumed to be merged in the disturbance
term, and they will not affect the statistical inferences much. On the other hand, if they are large in
magnitude, then they will lead to incorrect and invalid statistical inferences. For example, in the context of
linear regression model, the ordinary least squares estimator (OLSE) is the best linear unbiased estimator of
the regression coefficient when measurement errors are absent. When the measurement errors are present in
the data, the same OLSE becomes biased as well as inconsistent estimator of regression coefficients.

Consequences of measurement errors:

We first describe the measurement error model. Let the true relationship between correctly observed study
and explanatory variables be
y  X 

where y is a  n 1 vector of true observation on study variable, X is a  n  k  matrix of true observations

on explanatory variables and  is a  k  1 vector of regression coefficients. The value y and X are not

observable due to the presence of measurement errors. Instead, the values of y and X are observed with
additive measurement errors as
y  y  u
X  X  V
where y is a  n 1 vector of observed values of study variables which are observed with (n  1)

measurement error vector u . Similarly, X is a (n  k ) matrix of observed values of explanatory variables

which are observed with  n  k  matrix V of measurement errors in X . In such a case, the usual disturbance

term can be assumed to be subsumed in u without loss of generality. Since our aim is to see the impact of
measurement errors, so it is not considered separately in the present case.

Alternatively, the same setup can be expressed as

y  X   u
X  X  V
where it can be assumed that only X is measured with measurement errors V and u can be considered as the
usual disturbance term in the model.

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

2
In case, some of the explanatory variables are measured without any measurement error then the
corresponding values in V will be set to zero.

We assume that
E (u )  0, E (uu ')   2 I
E (V )  0, E (V 'V )  , E (V ' u )  0.

The following set of equations describes the measurement error model

y  X 
y  y  u
X  X  V
which can be re-expressed as
y  y  u
 X   u
  X V    u
 X   u V  
=X   
where   u  V  is called as the composite disturbance term. This model resemble like a usual linear
regression model. A basic assumption in linear regression model is that the explanatory variables and
disturbances are uncorrelated. Let us verify this assumption in the model y  X   w as follows:

E  X  E ( X ) '  E ( )  E V '(u  V  ) 

 E V ' u   E V 'V  
 0  
 
 0.
Thus X and  are correlated. So OLS will not provide efficient result.

Suppose we ignore the measurement errors and obtain the OLSE. Note that ignoring the measurement errors
in the data does not mean that they are not present. We now observe the properties of such an OLSE under
the setup of measurement error model.

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

3
The OLSE is

b X 'X  X 'y

1

b     X ' X  X ' X      
1

  X ' X  X '
1

E  b     E  X ' X  X '  
1
 
  X ' X  X ' E ( )
1

0
as X is a random matrix which is correlated with  . So b becomes a biased estimator of  .

Now we check the consistency property of OLSE. Assume

1 
plim  X ' X    xx
n 
1 
plim  V 'V    vv
n 
1 
plim  X 'V   0
n 
1 
plim  V ' u   0.
n 
Then
  X ' X  1  X '   
plim  b     plim    
 n   n  
1 1
n

X ' X  X  V ' X  V
n
 
1 1 1 1
 X ' X  X 'V  V ' X  V 'V
n n n n
1  1  1  1  1 
plim  X ' X   plim  X ' X   plim  X 'V   plim  V ' X   plim  V 'V 
n  n  n  n  n 
  xx  0  0   vv
  xx   vv

1 1 1
X '   X '   V ' 
n n n
1  1
 X '(u  V  )  V '(u  V  )
n n

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

4
 1  1  1  1  1 
plim  X '    plim  X ' u   plim  X 'V    plim  V ' u   plim  V 'V  
n  n  n  n  n 
 0  0  0  vv 
-1
 X 'X   X ' 
plim  b     plim   plim  
 n   n 
    xx  vv   vv 
1

 0.
Thus b is an inconsistent estimator of  . Such inconsistency arises essentially due to correlation between
X and  .

Note: It should not be misunderstood that the OLSE b   X ' X  X ' y is obtained by minimizing
1

S   '    y  X   '  y  X   in the model y  X    . In fact  '  cannot be minimized as in the case of

usual linear regression, because the composite error   u  V  is itself a function of  .

To see the nature of consistency, consider the simple linear regression model with measurement error as
yi   0  1 xi , i  1, 2,..., n
yi  yi  ui
xi  xi  vi .
Now
1 x1  1 x1   0 v1 
     
1 x2  1 x2  0 v2 
X  , X   , V 
        
     
1 xn  1 xn   0 vn 
and assuming that
1 n 
plim   xi   
 n i 1 
1 n 
plim   ( xi   ) 2    x2 ,
 n i 1 
we have

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

5
 1 
 xx  plim  X ' X 
n 
 1 n 
 1  xi
n i 1 
 plim  n
1 1 n 2
  xi  xi 
 n i 1 n i 1 
1  
 2
.
 x   
2

Also,
1 
 vv  plim  V 'V 
n 
0 0 
 2
.
0 v 
Now

plim  b        xx   vv   vv 
1

1
 b  0  1    0 0   0 
plim  0    2  2  
 b1  1     x     v   0  v   1 
2 2

1   x2   2   v2     0 
 2
 x   v2      2
1    v 
  v2 
 2 1 
v  x2

  .
  2 
  2 v 2 1 
  x v 

Thus we find that the OLSEs of  0 and 1 are biased and inconsistent. So if a variable is subjected to

measurement errors, it not only affects its own parameter estimate but also affect other estimator of
parameter that are associated with those variable which are measured without any error. So the presence of
measurement errors in even a single variable not only makes the OLSE of its own parameter inconsistent but
also makes the estimates of other regression coefficients inconsistent which are measured without any error.

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

6
Forms of measurement error model:
Based on the assumption about the true values of the explanatory variable, there are three forms of
measurement error model .

Consider the model

yi   0  1 xi , i  1, 2,..., n
yi  yi  ui
xi  xi  vi .

1. Functional form: When the xi ' s are unknown constants (fixed), then the measurement error model is

said to be in its functional form.

2. Structural form: When the xi ' s are identically and independently distributed random variables, say, with

mean  and variance  2  2  0  , the measurement error model is said to be in the structural form.

Note that in case of functional form,  2  0.

3. Ultrastructural form: When the xi ' s are independently distributed random variables with different

means, say i and variance  2  2  0  , then the model is said to be in the ultrastructural form. This form is

a synthesis of function and structural forms in the sense that both the forms are particular cases of
ultrastructural form.

Methods for consistent estimation of  :

The OLSE of  which is the best linear unbiased estimator becomes biased and inconsistent in the presence
of measurement errors. An important objective in measurement error models is how to obtain the consistent
estimators of regression coefficients. The instrumental variable estimation and method of maximum
likelihood (or method of moments) are utilized to obtain the consistent estimates of the parameters.

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

7
Instrumental variable estimation:
The instrumental variable method provides the consistent estimate of regression coefficients in linear
regression model when the explanatory variables and disturbance terms are correlated. Since in measurement
error model, the explanatory variables and disturbance are correlated, so this method helps. The instrumental
variable method consists of finding a set of variables which are correlated with the explanatory variables in
the model but uncorrelated with the composite disturbances, at least asymptotically, to ensure consistency.

Let Z1 , Z 2 ,..., Z k be the k instrumental variables. In the context of the model

y  X   ,   u  V  ,
let Z be the n  k matrix of k instrumental variables Z1 , Z 2 ,..., Z k , each having n observations such that

 Z and X are correlated, atleast asymptotically and

 Z and  are uncorrelated, at least asymptotically.

So we have
1 
plim  Z ' X    ZX
n 
1 
plim  Z '    0.
 n 

The instrumental variable estimator of  is given by

ˆIV   Z ' X  Z ' y

1

  Z ' X  Z ' X    
1

ˆIV     Z ' X  Z ' 

1

1


plim ˆIV    1  1 
 plim  Z ' X  plim  Z '  
n  n 
1
  ZX .0
 0.

So ˆIV is consistent estimator of  .

Any instrument that fulfils the requirement of being uncorrelated with the composite disturbance term and
correlated with explanatory variables will result in a consistent estimate of parameter. However, there can be
various sets of variables which satisfy these conditions to become instrumental variables. Different choices
of instruments give different consistent estimators. It is difficult to assert that which choice of instruments
Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur
8
will give an instrumental variable estimator having minimum asymptotic variance. Moreover, it is also
difficult to decide that which choice of the instrumental variable is better and more appropriate in
comparison to other. An additional difficulty is to check whether the chosen instruments are indeed
uncorrelated with the disturbance term or not.

Choice of instrument:
We discuss some popular choices of instruments in a univariate measurement error model. Consider the
model
yi   0  1 xi  i , i  ui  1vi , i  1, 2,..., n.

A variable that is likely to satisfy the two requirements of an instrumental variable is the discrete grouping
variable. The Wald’s, Bartlett’s and Durbin’s methods are based on different choices of discrete grouping
variables.

1. Wald’s method
Find the median of the given observations x1 , x2 ,..., xn . Now classify the observations by defining an

instrumental variable Z such that

1 if xi  median ( x1 , x2 ,..., xn )
Zi  
1 if xi  median ( x1 , x2 ,..., xn ).
In this case,
1 Z1  1 x1 
   
1 Z2  1 x2 
Z  , X  .
     
   
1 Zn  1 xn 

Now form two groups of observations as follows.

 One group with those xi ' s below the median of x1 , x2 ,..., xn . Find the means of yi ' s and xi ' s , say

y1 and x1 , respectively in this group..

 Another group with those xi ' s above the median of x1 , x2 ,..., xn . Find the means of yi ' s and

xi ' s, say y2 and x2 , respectively in this group.

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

9
 1 n 1 n
Now we find the instrumental variable estimator under this set up as follows. Let x  
n i 1
xi , y   yi .
n i 1

ˆIV   Z ' X  Z ' y

1

 n

 n  xi   n nx 

Z'X  n
i 1
   
n
 n
 0  x2  x1  
  Z i  Z i xi   2 
 i 1 i 1 
 n 
  yi   ny 
Z'y  n
i 1
 n  
    y2  y1  
  Z i yi   2 
 i 1 
1
 ˆ0 IV  n nx   ny 
  n
  
 ˆ  0  x2  x1    n  y2  y1  
 1IV   2  2 
 x2  x1  y 
2  2 x   

 x2  x1   0   y2  y1 
 1  2 
  y2  y1  
 y  x
  x2  x1  
 
  y2  y1  
 
  x2  x1  

y y
 ˆ1IV  2 1
x2  x1
 y2  y1 
ˆ0 IV  y   ˆ
 x  y  1IV x .
 x2  x 

If n is odd, then the middle observations can be deleted. Under fairly general conditions, the estimators are
consistent but are likely to have large sampling variance. This is the limitation of this method.

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

10
2. Bartlett’s method:
Let x1 , x2 ,..., xn be the n observations. Rank these observation and order them in an increasing or decreasing

order. Now three groups can be formed, each containing n / 3 observations. Define the instrumental variable
as
1 if observation is in the top group

Z i  0 if observation is in the middle group
1 if observation is in the bottom group.


Now discard the observations in the middle group and compute the means of yi ' s and xi ' s in

- bottom group, say y1 and x1 and

- top group, say y3 and x3 .

Substituting the values of X and Z in ˆIV   Z ' X  Z ' y and on solving, we get
1

y3  y1
ˆ1IV 
x3  x1
ˆ0 IV  y  ˆ1IV x .

These estimators are consistent. No conclusive pieces of evidence are available to compare Bartlett’s method
and Wald’s method but three grouping method generally provides more efficient estimates than two
grouping method is many cases.

3. Durbin’s method
Let x1 , x2 ,..., xn be the observations. Arrange these observations in an ascending order. Define the

instrumental variable Z i as the rank of xi . Then substituting the suitable values of Z and X in

ˆIV   Z ' X  Z ' y, we get the instrumental variable estimators

1

Z y  y
i i
ˆ 1IV  i 1
n
.
Z x  x
i 1
i i

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

11
When there are more than one explanatory variables, one may choose the instrument as the rank of that
particular variable.

Since the estimator uses more information, it is believed to be superior in efficiency to other grouping
methods. However, nothing definite is known about the efficiency of this method.

In general, the instrumental variable estimators may have fairly large standard errors in comparison to
ordinary least square estimators which is the price paid for inconsistency. However, inconsistent estimators
have little appeal.

Maximum likelihood estimation in structural form

Consider the maximum likelihood estimation of parameters in the simple measurement error model given by
yi   0  1 xi , i  1, 2,..., n
yi  yi  ui
xi  xi  vi .

Here  xi , yi  are unobservable and  xi , yi  are observable.

Assume
 2 if i  j
E  ui   0, E  ui u j    u
0 if i  j ,
 v2 if i  j
E  vi   0, E  vi v j   
0 if i  j,
E  uV
i j   0 for all i  1, 2,..., n; j  1, 2,..., n.

For the application of the method of maximum likelihood, we assume the normal distribution for ui and vi .

We consider the estimation of parameters in the structural form of the model in which xi ' s are stochastic. So

assume
xi ~ N   ,  2 

and xi ' s are independent of ui and vi .

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

12
Thus
E ( xi )  
Var ( xi )   2
E ( xi )  
Var  xi   E  xi  E ( xi ) 
2

 E  xi  vi   
2

 E  xi     E  vi2   2  xi    vi

2

  2   v2
E  yi    0  1 E  xi 
  0  1 .

Var  yi   E  yi  E ( yi ) 
2

 E   0   1 xi  ui   0  1 
2

 12 E  xi     E  ui2   21 E  xi    ui

2

 12 2   u2

Cov  xi , yi   E  xi  E ( xi ) yi  E ( yi )

 E  xi  vi   0  1 xi  ui   0  1
 1 E  xi     E  xi    ui  1 E  xi    vi  E  ui vi 
2

 1 2  0  0  0
 1 2 .
So

 yi    0  1   12 2   u2 1 2  

  ~  ,   .
   1 2
N
 xi    2   v2  
The likelihood function is the joint probability density function of ui and vi , i  1, 2,..., n as

L  f  u1u2 ,..., un , v1 , v2 ,..., vn 

 n 2  n 2
 1 
n /2
  i  u   vi 
 2 2 
exp   i 1
 exp   i 1 2 
 2 u  v    2 v 
2
 2 u 
   
 n 2  n 2

 1 
n /2
  i i  x  
x     yi   0  1 xi  
 2 2 
exp   i 1  exp   i 1 .
 2 u  v   2 v2
2
 2 u   
   
Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur
13
The log-likelihood is
n n

  xi  xi   y   0  1 xi 
2 2
i
L*  ln L  constant 
n
2
 ln  u2  ln  v2   i 1
2 u2
 i 1
2 v2
.

The normal equations are obtained by equating the partial differentiations equals to zero as
L * 1 n
(1)    yi  0  1 xi   0
 0  v2 i 1
L * 1 n
(2)   xi  yi  0  1 xi   0
1  v2 i 1
L * 1 
(3)  2  xi  xi   2  yi   0  1 xi   0, i  1, 2,..., n
xi  u v
L * n 1 n
4  i
x  xi 
2
(4)   
 u
2
2 u 2 v i 1
2

L * n 1 n
4  i
y   0  1 xi  .
2
(5)   
 v
2
2 v 2 v i 1
2

These are  n  4  equations in  n  4  parameters but summing equation (3) over i  1, 2,..., n and using

equation (4), we get

 v2  12 u2
which is undesirable.

These equations can be used to estimate the two means   and  0  1  , two variances and one covariance.

The six parameters  ,  0 , 1 ,  u2 ,  v2 and  2 can be estimated from the following five structural relations
derived from these normal equations

(i ) x 
(ii ) y   0  1
(iii) mxx   2   v2
(iv) m yy  12 2   u2
(v ) mxy  1 2

1 n 1 n 1 n 1 n 1 n
i  i xx n     i  ( xi  x )( yi  y ).
2
where x  x , y  y , m  xi  x , m yy  ( y  y ) 2
and m xy 
n i 1 n i 1 i 1 n i 1 n i 1

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

14
These equations can be derived directly using the sufficiency property of the parameters in bivariate normal
distribution using the definition of structural relationship as

E ( x)  
E ( y )   0  1
Var ( x)   2   v2
Var ( y )  12 2   u2
Cov( x, y )  1 2 .

We observe that there are six parameters  0 , 1 ,  ,  2 ,  u2 and  v2 to be estimated based on five structural

equations (i)-(v). So no unique solution exists. Only  can be uniquely determined while remaining
parameters can not be uniquely determined. So only  is identifiable and remaining parameters are
unidentifiable. This is called the problem of identification. One relation is short to obtain a unique solution,
so additional a priori restrictions relating any of the six parameters is required.

Note: The same equations (i)-(v) can also be derived using the method of moments. The structural
equations are derived by equating the sample and population moments. The assumption of normal
distribution for ui , vi and xi is not needed in case of method of moments.

Additional information for the consistent estimation of parameters:

The parameters in the model can be consistently estimated only when some additional information about the
model is available.

From equations (i) and (ii), we have

ˆ  x
and so  is clearly estimated. Further

ˆ0  y  ˆ1 x

is estimated if ̂1 is uniquely determined. So we consider the estimation of 1 ,  2 ,  u2 and  v2 only. Some

additional information is required for the unique determination of these parameters. We consider now
various type of additional information which are used for estimating the parameters uniquely.

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

15
1.  v2 is known:

Suppose  v2 is known a priori. Now the remaining parameters can be estimated as follows:

mxx   2   v2  ˆ 2  mxx   v2
mxy
mxy  1 2  ˆ1 
mxx   v2
m yy  1 2   u2  ˆ u2  myy  ˆ12ˆ 2
mxy2
 myy  .
mxx   v2

Note that ˆ 2  mxx   v2 can be negative because  v2 is known and mxx is based upon sample. So we assume

that ˆ 2  0 and redefine

mxy
ˆ1  ; mxx   v2 .
mxx   2
v

Similarly, ˆ u2 is also assumed to be positive under suitable condition. All the estimators ˆ1 , ˆ 2 and ˆ u2 are

the consistent estimators of  ,  2 and  u2 respectively. Note that ˆ1 looks like as if the direct regression

estimator of 1 has been adjusted by  v2 for its inconsistency. So it is termed as adjusted estimator also.

2.  u2 is known
Suppose  u2 is known a priori. Then using mxy  1 2 , we can rewrite

m yy  12 2   u2
 mxy 1   u2
m yy   u2
 ˆ1  ; myy   u2
mxy
mxy
ˆ 2 
ˆ1

ˆ  mxx  ˆ 2 .
2
v

The estimators ˆ1 , ˆ 2 and ˆ v2 are the consistent estimators of 1 ,  2 and  v2 respectively. Note that ̂1 looks

like as if the reverse regression estimator of 1 is adjusted by  u2 for its inconsistency. So it is termed as

adjusted estimator also.

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur
16
  u2
3.   is known
 v2
Suppose the ratio of the measurement error variances is known, so let
 u2

 v2
is known.

Consider
m yy  12 2   u2
 1 mxy   v2 (using (iv))
 1 mxy    mxx   2  (using (iii))
 m 
 1 mxy    mxx  xy  (using iv)
 1 

12 mxy    1mxx  mxy   1myy  0  1  0 

12 mxy     mxx  m yy    mxy  0.

Solving this quadratic equation

m   mxx   m   mxx   4 mxy2

2

ˆ1 
yy yy

2mxy
U
 , say.
2mxy

Since mxy  1 2 and

ˆ 2  0
mxy
 0
ˆ1

2mxy2
 0
U
 since mxy2  0, so U must be nonnegative.

This implies that the positive sign in U has to be considered and so

m   mxx   m   mxx   4 mxy2

2

ˆ1 
yy yy
.
2mxy
Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur
17
Other estimates are
myy  2ˆ1mxy  ˆ12 sxx
ˆ v 
2

  ˆ 2 1

mxy
ˆ 2  .
ˆ1

Note that the same estimator ˆ1 of 1 can be obtained by orthogonal regression. This amounts to transform

xi by xi /  u and yi by yi /  v and use the orthogonal regression estimation with transformed variables.

4. Reliability ratio is known

The reliability ratio associated with the explanatory variable is defined as the ratio of variances of true and
observed values of explanatory variables, so
Var ( x ) 2
Kx   2 ; 0  Kx  1
Var ( x)    v2
is the reliability ratio. Note that K x  1, when  v2  0 which means that there is no measurement error in the

explanatory variable and K x  0, means  2  0 which means the explanatory variable is fixed. A higher

value of K x is obtained when  v2 is small, i.e., the impact of measurement errors is small. The reliability

ratio is a popular measure in psychometrics.

Let K x be known a priori. Then

mxx   2   v2
mxy  1 2
mxy 1 2
 
mxx  2   v2
 1 K x
mxy
 ˆ1 
K x mxx
m
 2  xy
1
 ˆ 2  K x mxx
mxx   2   v2
 ˆ v2  1  K x  mxx .
Note that ˆ  K 1b
1 x

mxy
where b is the ordinary least squares estimator b  .
mxx
Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur
18
5.  0 is known
Suppose  0 is known a priori and E ( x)  0. Then

y   0  1
y  0
 ˆ1 
ˆ
y  0

x
m
ˆ 2  xy
ˆ 1

ˆ  myy  ˆ1mxy
2
u

mxy
ˆ v2  mxx  .
ˆ 1

6. Both  u2 and  v2 are known

This case leads to over-identification in the sense that the number of parameters to be estimated is smaller
than the number of structural relationships binding them. So no unique solutions are obtained in this case.

Note: In each of the cases 1-6, note that the form of the estimate depends on the type of available
information which is needed for the consistent estimator of the parameters. Such information can be
available from various sources, e.g., long association of the experimenter with the experiment, similar type
of studies conducted in the part, some external source etc.

Estimation of parameters in function form:

In the functional form of the measurement error model, xi ' s are assumed to be fixed. This assumption is

unrealistic in the sense that when xi ' s are unobservable and unknown, it is difficult to know if they are fixed

or not. This can not be ensured even in repeated sampling that the same value is repeated. All that can be said
in this case is that the information, in this case, is conditional upon xi ' s . So assume that xi ' s are

conditionally known. So the model is

yi   0  1 xi
xi  xi  vi
yi  yi  ui
then

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

19
  yi    0  1 xi    u2 0  
  ~ N  , 2 
.
 xi   xi   0  v  

The likelihood function is

 n 2

 1    n
2
 yi   0  1 xi  
 1 
n
2    xi  xi 2 
L 2 
exp   i 1
 2 
exp   .
 2 u   2 u2   2 v   2 v2 
 

The log-likelihood is
n

 y    1 xi 
2
0
n i
n 1 n

  x  x  .
2
L*  ln L  constant  ln  u2  i 1
 ln  v2  2
2 2
2 v
i i
2 u 2 i 1

The normal equations are obtained by partially differentiating L * and equating to zero as
L * 1 n
(I )
 0
0 2
u
 y  
i 1
i 0  1 xi   0

L * 1 n
( II )
1
0 2
v
 y  
i 1
i 0  1 xi  xi  0

L * n 1 n

 y    1 xi   0
2
( III ) 0 2  4
 u 2 u 2 u
2 i 0
i 1

L * n 1 n

  x  x 
2
( IV ) 0 2  4 0
 v
2
2 v 2 v
i i
i 1

L *  1
(V )  0  2  yi   0  1 xi   2 ( xi  xi )  0.
xi u v

Squaring and summing equation (V), we get

2 2
   1 
i  12  yi  0  1 xi   i    2  xi  xi 
 u   v 
 2
1
or 12   yi   0  1 xi   4   xi  xi  .
2 2

u i v i

Using the left-hand side of equation (III) and right-hand side of equation (IV), we get
n12 n

 2
u  v2
u
 1 
v
Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur
20
which is unacceptable because  can be negative also. In the present case, as  u  0 and  v  0 , so  will

always be positive. Thus the maximum likelihood breaks down because of insufficient information in the
model. Increasing the sample size n does not solve the purpose. If the restrictions like  u2 known,  v2

 u2
known or known are incorporated, then the maximum likelihood estimation is similar to as in the case of
 v2

structural form and the similar estimates may be obtained. For example, if    u2 /  v2 is known, then

substitute it in the likelihood function and maximize it. The same solution as in the case of structural form
are obtained.

Econometrics | Chapter 16 | Measurement Error Models | Shalabh, IIT Kanpur

21