Scribd
Upload
328 views

Multiple Regression Analysis: The Problem of Estimation: Gujarati 5e, Chapter 7

The document discusses the assumptions and properties of multiple linear regression analysis. It describes a basic three-variable regression model and defines the parameters. It then lists the classical linear regression model assumptions, such as linearity, zero conditional mean, homoscedasticity, no autocorrelation or multicollinearity. It also explains how to estimate the regression coefficients using ordinary least squares and defines terms like the partial regression coefficients, R-squared, and adjusted R-squared.

Uploaded by

Raditya
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
328 views

Multiple Regression Analysis: The Problem of Estimation: Gujarati 5e, Chapter 7

The document discusses the assumptions and properties of multiple linear regression analysis. It describes a basic three-variable regression model and defines the parameters. It then lists the classical linear regression model assumptions, such as linearity, zero conditional mean, homoscedasticity, no autocorrelation or multicollinearity. It also explains how to estimate the regression coefficients using ordinary least squares and defines terms like the partial regression coefficients, R-squared, and adjusted R-squared.

Uploaded by

Raditya
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 20

Multiple regression analysis:

the problem of estimation

Gujarati 5e, Chapter 7


The three-variable model

Yi  1   2 X 2i  3 X 3i  ui
where:
Y is the dependent variable
X2 and x3 are the explanatory variables
u is the stochastic disturbance term
1 is the intercept term
2 and 3 are the partial regression coefficients

09/08/20 Prepared by Sri Yani K 2


CLRM Assumption (1)

1. Linear regression model, or linear in the


parameters.
2. Fixed X values or X values independent of the
error term. Here, this means we require zero
covariance between ui and each X variables.
cov (ui , X2i) = cov (ui , X3i) = 0
n Zero mean value of disturbance ui.
E(ui | X2i, X3i) = 0 for each i
n Homoscedasticity or constant variance of ui.
var (ui ) = σ2

09/08/20 Prepared by Sri Yani K 3


CLRM Assumption (2)

5. No autocorrelation, or serial correlation,


between the disturbances.
cov (ui , uj) = 0 i=j
6. The number of observations n must be
greater than the number of parameters to be
estimated, which is 3 in our current case.
7. There must be variation in the values of the
X variables.

09/08/20 Prepared by Sri Yani K 4


CLRM Assumption (3)

8. No exact collinearity between the X


variables. No exact linear relationship
between X2 and X3
9. There is no specification bias. The model is
correctly specified

09/08/20 Prepared by Sri Yani K 5


Interpretation of multiple regression
equation
E  Yi X 2i , X 3i   1   2 X 2i  3 X 3i

 The conditional mean or expected value of Y


conditional upon the given or fixed values of
the variables X2 and X3
 The average mean value of Y or mean
response of Y for the fixed values of the X
variables

09/08/20 Prepared by Sri Yani K 6


The meaning of partial regression
coefficients
 2 measures the change in the mean value of
Y, E(YX2i, X3i), per unit change in X2,
holding X3 constant
 3 measures the change in the mean value of
Y per unit change in X3, holding X2 constant

09/08/20 Prepared by Sri Yani K 7


OLS estimation of partial regression
coefficients
 
2
 SRF: min  uˆ i  Yi  ˆ1  ˆ2 X 2i  ˆ3 X 3i
2

Yi  ˆ1  ˆ2 X 2i  ˆ3 X 3i  uˆi


 The normal equations:
Yi  ˆ1  ˆ2 X 2  ˆ3 X 3i
 Y X  ˆ
i 2i 
1X  ˆ 2i 2 X 2
2i  ˆ
 3  X 2 i X 3i

Y X i 3i  ˆ  X
1 3i  ˆ
 2 X X
2 i 3i  ˆ
 3 X 2
3i

09/08/20 Prepared by Sri Yani K 8


OLS estimators
ˆ1  Y  ˆ2 X  ˆ3 X 3

  y x   x    y x    x
2
x 
ˆ2
i 2i 3i i 3i 2 i 3i

 x   x    x x 
2 2 2
2i 3i 2 i 3i

ˆ3 
  yi x3i   
 2i    yi x2i    x2i x3i 
x 2

  x    x    x 
2 2 2
2i 3i x
2 i 3i

09/08/20 Prepared by Sri Yani K 9


Variance of OLS estimators
 
  1
var ˆ1    2
X 2
 x 2
3i  X 2
3  x 2
2i  2 X 2 X 3  x x
2 i 3i
 2
n      
2

 x2
2
i x3
2
i  x 2 i x3 i 

 
var ˆ2 
 3ix 2

 2

 x2i  x3i    x2i x3i 


2 2 2


var ˆ3
 x
 x x  x
2
2i
 2

   x3i 
2 2 2
2i 3i 2i

 r23 2

cov ˆ2 , ˆ3  
 1 r 
2
23 x22i x32i

2   i
ˆ
u 2

n3

09/08/20 Prepared by Sri Yani K 10


Properties of OLS estimators

1. The regression line passes through the mean


Y , X 21 , Xand
31
2. The mean value of estimated is equal to the
mean value of actual
3. The sum of residual is equal the mean value
of residual and equal zero:  uˆi  u  0
4. The residual are uncorrelated with X2i and X3i:
5.  uˆi X 2i are
The residual uncorrelated
uˆi X 3i  0 with
Yˆi
 uˆiYˆi  0
09/08/20 Prepared by Sri Yani K 11
Properties of OLS estimators
6. r23, the coefficient correlation between X2i and
X3i, increase toward 1, the variance of
coefficient increase for given value of 2 and
 2 i  3i
x 2
or x 2

For given value of r23 and  2i  3,i the


2 2
7.
x or x
variance of the OLS estimators are directly
proportional to 2
8. Given the assumption of the CLRM, they are
BLUE

09/08/20 Prepared by Sri Yani K 12


The multiple coefficient of
determination, R2
 R2 measures the goodness of fit the regression
equation  the proportion of the variation in Y
explained by the variables X2 and X3 jointly

2  i 2i 3  i 3i
ESS ˆ
 y x  ˆ yx
R 
2

TSS  yi 2

2  

 
var ˆ j 
x 2 
 j j
 1 
1
R 2 

09/08/20 Prepared by Sri Yani K 13


R2 and the adjusted R2

 An important properties of R2 is that it is a


nondecreasing function of the number of
explanatory variables.

R 
2 ESS
 1
RSS
 1
 ˆ
ui
2

TSS TSS  yi2

 The adjusted R2

R2  1
 i  nk
ˆ
u 2
 Adjusted for the df
 i  n  1
y 2

09/08/20 Prepared by Sri Yani K 14


R2 and the adjusted R2
n 1
2

R  1 1 R 2
 nk
1. For k>1, adjusted R2 < R2 which implies that as the
number of X variables increases, the adjusted R2
increases less than the unadjusted R2
2. Adjusted R2 can be negative

 Other criteria: Akaike’s Information Criterion


(AIC) and Amemiya’s Precidtion criterion

09/08/20 Prepared by Sri Yani K 15


Comparing the two R2 values
 In comparing two models on the basis of R2 or
adjusted R2: the sample size n and the
dependent variable must be the same; the
explanatory variables may take any form
 For the models
ln Yi  1   2 X 2i  3 X 3i  ui
Yi  1   2 X 2i   3 X 3i  ui
R2 cannot be compared

09/08/20 Prepared by Sri Yani K 16


The “game” of maximizing R2
 Sometimes researcher play the game of
maximizing adjusted R2, that is, choosing the
model that gives the highest adjusted R2. But
this may be dangerous.
 The researcher should be more concerned
about the logical or theoretical relevance of the
explanatory variables to the dependent variable
and their statistical significance.
 If adjusted R2 is low, it does not mean the model
is necessarily bad.

09/08/20 Prepared by Sri Yani K 17


The Cobb-Douglas production
function
2 3 ui
Yi  1 X X e 2i 3i
where Y=output
X2=labor input
X3=capital input
u=stochastic disturbance term
e=base of natural logarithm
 Model transform:
ln Yi   0   2 ln X 2i   3 ln X 3i  ui
where  0  ln 1
09/08/20 Prepared by Sri Yani K 18
The properties of the Cobb-Douglas
production function
1. 2 is the partial elasticity of output with respect
to the labor input, holding the capital input
constant
2. 3 is the partial elasticity of output with respect
to the capital input, holding the labor input
constant
3. The sum (2+3) gives information about the
returns to scale, that is, the response of output
to a proportionate in the inputs

09/08/20 Prepared by Sri Yani K 19


Polynomial regression models

 The general kth degree polynomial regression


may be written as

Yi   0  1 X i   2 X  ...   k X  ui
i
2
i
k

 ’s can be estimated by the OLS or ML


methodology

09/08/20 Prepared by Sri Yani K 20

You might also like