0% found this document useful (0 votes)
52 views26 pages

Chapter 8A

Uploaded by

arianagacosta
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
52 views26 pages

Chapter 8A

Uploaded by

arianagacosta
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 26

Chapter 8 – Part A

Nonlinear Regression Functions

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


Outline

• Modeling nonlinear regression functions (Ch. 8.1)


– The effect on Y of a change in X in nonlinear
specifications
• Nonlinear functions of a single independent variable (Ch.
8.2)
– Polynomials
– Logarithms

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-2
TestScore – Income relation looks
nonlinear...

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-3
Nonlinear Regression Population

Yi = f(X1i, X2i,…, Xki) + ui, i = 1,…, n

If a relation between Y and X is nonlinear:


• The effect on Y of a change in X depends on the value of X – that
is, the marginal effect of X is not constant.

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-4
Nonlinear Functions of a Single
Independent Variable

Two complementary approaches:

1. Polynomials in X
– The population regression function is approximated by a
quadratic, cubic, or higher-degree polynomial

2. Logarithmic transformations
– Y and/or X is transformed by taking its logarithm
– “Percentages” interpretation

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-5
1. Polynomials in X

Approximate the population regression function by a polynomial:

Yi = β0 + β1Xi + β2 X2i +…+ βr Xri + ui

• This is just the linear multiple regression model – except that the
regressors are powers of X!
• 2 regressors: quadratic regression
• 3 regressors: cubic regression, etc.
• Estimation, hypothesis testing, etc. proceeds as in the multiple
regression model using OLS

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-6
Example: the TestScore – Income
relation

Quadratic specification:

TestScorei = β0 + β1Incomei + β2(Incomei)2 + ui

Cubic specification:

TestScorei = β0 + β1Incomei + β2(Incomei)2 + β3(Incomei)3 + ui

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-7
Estimation of the quadratic specification
generate avginc2 = avginc*avginc; Create a new regressor
reg testscr avginc avginc2, r;

Regression with robust standard errors Number of obs = 420


F( 2, 417) = 428.52
Prob > F = 0.0000
R-squared = 0.5562
Root MSE = 12.724

------------------------------------------------------------------------------
| Robust
testscr | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
avginc | 3.850995 .2680941 14.36 0.000 3.32401 4.377979
avginc2 | -.0423085 .0047803 -8.85 0.000 -.051705 -.0329119
_cons | 607.3017 2.901754 209.29 0.000 601.5978 613.0056
------------------------------------------------------------------------------

Test the null hypothesis of linearity: H0: β2=0

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-8
Interpreting the estimated regression
function (I):
(I) Plot the predicted values

TestScore = 607.3 + 3.85Incomei – 0.0423(Incomei)2
(2.9) (0.27) (0.0048)

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-9
Interpreting the estimated regression
function (II) :

(b) Compute “effects” for different values of X


TestScore = 607.3 + 3.85Incomei – 0.0423(Incomei)2
(2.9) (0.27) (0.0048)

Predicted change in TestScore for a change in income from


$5,000 per capita to $6,000 per capita:


Δ TestScore = 607.3 + 3.85×6 – 0.0423×62
– (607.3 + 3.85×5 – 0.0423×52)
= 3.4

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-10

TestScore = 607.3 + 3.85Incomei – 0.0423(Incomei)2

Predicted “effects” for different values of X:


ΔTestScore
Change in Income ($1000 per capita)

from 5 to 6 3.4

from 25 to 26 1.7

from 45 to 46 0.0

• The “effect” of a change in income is greater at low than high


income levels.
• Remember not to extrapolate out of the sample.

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-11
Estimation of a cubic specification

gen avginc3 = avginc*avginc2; Create the cubic regressor


reg testscr avginc avginc2 avginc3, r;

Regression with robust standard errors Number of obs = 420


F( 3, 416) = 270.18
Prob > F = 0.0000
R-squared = 0.5584
Root MSE = 12.707

------------------------------------------------------------------------------
| Robust
testscr | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
avginc | 5.018677 .7073505 7.10 0.000 3.628251 6.409104
avginc2 | -.0958052 .0289537 -3.31 0.001 -.1527191 -.0388913
avginc3 | .0006855 .0003471 1.98 0.049 3.27e-06 .0013677
_cons | 600.079 5.102062 117.61 0.000 590.0499 610.108
------------------------------------------------------------------------------

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-12
Testing for linearity
Testing the null hypothesis of linearity:

H0: population coefficients on Income2 and Income3 = 0


H1: at least one of these coefficients is nonzero.

test avginc2 avginc3; Execute the test command after running the regression

( 1) avginc2 = 0.0
( 2) avginc3 = 0.0

F( 2, 416) = 37.69
Prob > F = 0.0000

The hypothesis that the population regression is linear is rejected at


the 1% significance level against the alternative that it is a polynomial
of degree up to 3.

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-13
Which degree polynomial to use?

• Trade-off between flexibility (a degree r polynomial can have up


to r-1 bends) and statistical precision (more coefficients to be
estimated)
1. Pick a maximum starting value for r.
2. Test of coefficient on Xr=0
3. If you reject H0 à polynomial of degree r
4. If you do not reject H0 à estimate a polynomial of degree r-1
5. Loop

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-14
2. Logarithmic functions of Y and/or X

• ln(X) = the natural logarithm of X


• Logarithm is the inverse of exponential function: X=ln(ex)
• Logarithmic transforms permit modeling relations in “percentage”
terms, rather than linearly.

æ Dx ö Dx
Here’s why: ln(x+Δx) – ln(x) = ln 1+ ≅
çè ÷
x ø x
Numerically:
ln(101)-ln(100) = ln(1+0.01) = ln(1.01) = .00995 ≅ .01
Similarly ln(1.10) = .0953 ≅ .10

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-15
The 3 log-regression specifications:

Case Population regression function

I. linear-log Yi = β0 + β1ln(Xi) + ui
II. log-linear ln(Yi) = β0 + β1Xi + ui
III. log-log ln(Yi) = β0 + β1ln(Xi) + ui

The interpretation of the slope coefficient differs in each case.

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-16
I. Linear-log regressions
Compute Y “before” and “after” changing X:
Y = β0 + β1ln(X) (“before”)

Now change X: Y + ΔY = β0 + β1ln(X + ΔX) (“after”)

Subtract (“after”) – (“before”): ΔY = β1[ln(X + ΔX) – ln(X)]


DX
now ln(X + ΔX) – ln(X) ≅ ,
X
DX
So ΔY ≅ β1
X
DY
or β1 ≅ (for small ΔX)
DX / X
• Interpretation: a 100% change in X is associated with a β1
change in Y.
DX
• Interpretation: a 1% change in X (0.01 increase in )
X
implies Y to increase by 0.01 β1 units.
Copyright © 2011 Pearson Addison-Wesley. All rights reserved.
8-17
Example: TestScore vs. ln(Income)

• First defining the new regressor, ln(Income)


• The linear-log model can be estimated by OLS:

= 557.8 + 36.42×ln(Incomei)
TestScore
(3.8) (1.40)

so a 1% increase in Income is associated with an increase in


TestScore of 0.01*36.42 = 0.36 points on the test.
• Standard errors, confidence intervals, R2 – all the usual tools of
regression apply here.
• How does this compare to the cubic model?

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-18
The linear-log and cubic regression
functions

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-19
II. Log-linear regressions

ln(Y) = β0 + β1X (“before”)

Now change X: ln(Y + ΔY) = β0 + β1(X + ΔX) (“after”)

Subtract (a) – (b): ln(Y + ΔY) – ln(Y) = β1ΔX


DY
so ≅ β1ΔX
Y
DY / Y
or β1 ≅ (small ΔX)
DX

• Interpretation: a 1 unit increase in X implies Y to increase by


(100*β1)%.

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-20
Example: ln(Earnings) vs. Age

• The log-linear model:


Earnings = 2.811 + 0.0096×Agei

so a 1 unit increase in Age is associated with an increase in


Earnings of 0.0096x100% à 0.96%.

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-21
III. Log-log regressions
ln(Yi) = β0 + β1ln(Xi) + ui (“before”)

Now change X: ln(Y + ΔY) = β0 + β1ln(X + ΔX) (“after”)

Subtract a-b: ln(Y + ΔY) – ln(Y) = β1[ln(X + βX) – ln(X)]

so
DY ≅ β DX
1
Y X
DY / Y
or β1 ≅ (small ΔX)
DX / X

• Interpretation: a 1% increase in X causes a β1% increase in Y


à elasticity

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-22
Example: ln(TestScore) vs. ln(Income)

• The model:


ln(TestScore ) = 6.336 + 0.0554×ln(Incomei)
(0.006) (0.0021)

An 1% increase in Income is associated with an increase of


0.0554% in TestScore.

An 100% increase in Income is associated with an increase of


5.54% in TestScore.

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-23
Example: ln( TestScore) vs. ln( Income)


ln(TestScore ) = 6.336 + 0.0554×ln(Incomei)
(0.006) (0.0021)

• For example, suppose income increases from $10,000 to $11,000,


or by 10%. Then TestScore increases by approximately .0554×10
= 0.554%.

• If TestScore = 650, this corresponds to an increase of


0.00554×650 = 3.6 points.

• How does this compare to the log-linear model?

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-24
The log-linear and log-log regressions:

• Notice vertical axis

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-25
Comparing R2 across specifications

• R2 can be used to compare:


– Linear vs linear-log models
– Log-linear vs log-log models
• But not to compare:
– Linear-log vs log-linear or log-log models
• The outcome variable needs to be the same.

Copyright © 2011 Pearson Addison-Wesley. All rights reserved.


8-26

You might also like