Scribd
Upload
41 views14 pages

Functional Form and Scaling: - Objectives

1) The lecture discusses functional forms and scaling of variables in regression analysis. 2) It explains how to interpret coefficients and elasticities in linear, double-log, and other functional forms. 3) A key point is that changing the units of measurement does not affect conclusions like t-values and R-squared in linear regressions, or the elasticity coefficient in double-log models. However, the intercept term and its t-value can be unit-dependent in double-log models.

Uploaded by

Syaharani
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
41 views14 pages

Functional Form and Scaling: - Objectives

1) The lecture discusses functional forms and scaling of variables in regression analysis. 2) It explains how to interpret coefficients and elasticities in linear, double-log, and other functional forms. 3) A key point is that changing the units of measurement does not affect conclusions like t-values and R-squared in linear regressions, or the elasticity coefficient in double-log models. However, the intercept term and its t-value can be unit-dependent in double-log models.

Uploaded by

Syaharani
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 14

Lecture 5

Functional Form and Scaling


• Objectives
– By the end of this section students will
• Understand how to interpret the coefficients from
regressions in a variety of functional forms
• Understand why it does not matter in which units
variables are measured
• References
– Gujarati, pp169-191
Functional Form
• Marginal
– The response of Y to a small change in X i.e.
dY/dX
• Elasticity
– The proportionate response of Y to X
dY dX dY X
elasticity  / 
Y X dX Y
marginal

average
Functional Form
• Responses are usually evaluated at the
mean of Y and X
• Example
– Given the following results calculate two
measures of the elasticity of housing prices
with respect to distance from the GPO.
Y  213.16 and X  11.25
C
o
mmo
nFu
nc
t
io
na
lF
or
ms

M
a
rg
i
nal E
l
ast
ici
ty
1
)L
in
ea
r
X
Y
=
+X  
()
Y

2
)Do
ub
l
e-L
og

Y
l
ogY
=
+l
ogX 
( ) 
X

3
)S
em
i-
Log
:l
ine
ar
-l
og

 
Y
=
+l
ogX X Y
4) Semi-Log: log-linear

logY =  + X Y X

5) Inverse

1  
Y =  +  (X ) - 2 - XY
X

6) Quadratic

2X2
Y =  +  X2 2X
Y
Functional Forms
• Estimation Results
Y  272.84  5.304 X R 2  0.4646
(15.96) (1.139)
log Y  5.985  0.2714 log X R 2  0.6039
(0.100) (0.0440)
X
• Linear model elasticity    
Y 
 5.30411 .25 / 213.16   0.2799
Functional Form
• Double Log Model
– Elasticity = = -0.2714
• Note you cannot compare R2 across the two
models - the dependant variables differ
– Linear model: Y
– Double Log model: log(Y)
Functional Form
• Non-linear Least Squares
– If a functional form is appropriate which is not
linear in the coefficients then OLS cannot be
used to estimate the coefficients. In this case
non-linear least squares (NLS) must be used.
Non-linear least squares is beyond the scope of
this course. NLS estimators are not BLUE.

Y  X
Scaling of Units
• Example
– Linear Model
– Does it matter if we measure X (distance from
GPO) in metres rather that kilometres?
X in kilometres: Y = 272.84 - 5.304 X R2 = 0.4646
(15.96) (1.139) s = 49.452

X in metres Y = 272.84 - 0.005304 X R2 = 0.4646


(15.96) (0.001139) s = 49.452
Scaling of Units
• Does it make any difference if we measure Y
(house prices) in dollars rather than
thousands of dollars?
Y in $’000 Y = 272.84 - 5.304 X R2 = 0.4646
(15.96) (1.139) s = 49.452

Y in dollars Y = 272840 -5304 X R2 = 0.4646


(15960) (1139) s = 49452
Scaling of Units
• Linear Model
– t-values and R2 do not change as the units
change
  and s depend on the units in which Y is
measured
  depends on the units of measurement of both
X and Y
Scaling of Units
• Double Log Model
Y in $’000 logY = 5.895 - 0.2714 logX R2 = 0.6039
(0.100) (0.0440)

Y in dollars logY = 12.803 - 0.2714 logX R2 = 0.6039


(0.100) (0.0440)
Scaling of Units
• Double Log Model
  is unit free (as you would expect because it is
an elasticity).
– The t-value of  and the R2 are unit free.
–  and its t-value are not unit-free. This follows
from the property of logs:
log(kY )  log k  log Y
Summary and Conclusions
• Functional form can affect interpretation of
parameters
• The units of measure never affect the conclusions
you make from your estimation results
• Next lecture: Meaning of Intercept & Multiple
Regression
– Gujarati, pp. 164-169, 202-207, 210-215, 217-223

You might also like