Topic 5

Functional Forms and

Qualitative Regressors

Pham Nhu Man

[email protected]
OUTLINE
Functional forms
• Linear
• Log-linear
• Log-lin
• Lin-log
Quadratic functional form
Dummy variable as a regressor
Transforming categorical variables into dummies
Dummy regressor and structural change/difference
LINEAR AND OTHER FUNCTIONAL FORMS

» Consider a linear regression function

𝑌 = 𝛽0 + 𝛽1 𝑋 + 𝜀
» 𝛽 is change in Y when X increases by 1 unit.
» Sometimes the relationship is not linear.
» Common functional form:
• Linear
• Log-linear
• Log-lin
• Lin-log

3
FUNCTIONAL FORMS

Linear Log-linear

Lin-log Log-lin

4
EXAMPLE DATASET

Viet Nam Provincial data on (file ‘fnf.xlsx’)

» gdp: provincial GDP (bil. VND)
» labfo: number of laborers of provinces (1000 persons)
» rinvest: gross investment of provinces (bil. VND)
» year: 2007 - 2011
» data for 58 provinces, 5 years (2007-2011)
» The data has some missing values.
DATA DESRIPTION

6
DATA DESCRIPTION

7
DATA DESCRIPTION

8
 EXPLAINING THE LINEAR MODEL
» Consider a linear regression function

𝑌 = 𝛽0 + 𝛽1 𝑋 + 𝜀
» Slope (marginal effect): 𝛽1
When 𝑋 increases by 1 unit, 𝑌 changes by 𝛽1 .

𝑋
» Elasticity: 𝛽1
𝑌
𝑋
When 𝑋 increase by 1%, 𝑌 changes by 𝛽1 percent.
𝑌

So elasticity of a linear model depends on the values of 𝑋 and 𝑌.

REGRESSION: LINEAR MODEL

10
 LOG-LINEAR MODEL

» The Cobb-Douglas Production Function:

𝑄𝑖 = 𝛽1 𝐿𝑖 𝛽2 𝐾𝑖 𝛽3

can be transformed into a linear model by taking natural logs of both sides:
ln 𝑄𝑖 = ln 𝛽1 + 𝛽2 ln 𝐿𝑖 + 𝛽3 ln 𝐾𝑖

» The slope coefficients can be interpreted as elasticities.

» Particularly for this production function:
➢ If (β2 + β3) = 1 : constant returns to scale.
➢ If (β2 + β3) > 1 : increasing returns to scale.
➢ If (β2 + β3) < 1 : decreasing returns to scale.
 EXPLAINING THE LOG-LINEAR MODEL

» Consider a linear regression function

ln 𝑌 = 𝛽0 + 𝛽1 ln 𝑋 + 𝜀

𝑌
» Slope (marginal effect): 𝛽1
𝑋
𝑌
When 𝑋 increases by 1 unit, 𝑌 changes by 𝛽1 .
𝑋
The marginal effect in a log-linear model depends on the values of 𝑋 and 𝑌.

» Elasticity: 𝛽1
When 𝑋 increase by 1%, 𝑌 changes by 𝛽1 percent.
LOG-LINEAR MODEL

13
LOG-LIN MODEL (GROWTH MODEL)

» The rate of growth of real GDP

𝑅𝐺𝐷𝑃𝑡 = 𝑅𝐺𝐷𝑃0 (1 + 𝑟)𝑡

can be transformed into a linear model by taking natural logs of both sides:

ln 𝑅 𝐺𝐷𝑃𝑡 = ln 𝑅 𝐺𝐷𝑃0 + 𝑡 ln( 1 + 𝑟)

» Let B0 = lnRGDP0 and B1 = ln(l+r), this can be rewritten as:

ln 𝑅𝐺𝐷𝑃𝑡 = 𝛽0 + 𝛽1 𝑡
 EXPLAINING THE LOG-LIN MODEL

» The model:
ln 𝑌 = 𝛽0 + 𝛽1 𝑋 + 𝜀
» Slope (marginal effect): 𝛽1 𝑌
When 𝑋 increases by 1 unit, 𝑌 changes by 𝛽1 𝑌.
The marginal effect in a log-linear model depends on the current value of 𝑌.

» Elasticity: 𝛽1 𝑋
When 𝑋 increase by 1%, 𝑌 changes by 𝛽1 𝑋 percent.
The elasticity in a log-linear model depends on the current value of 𝑋.

» The meaning of β1 : Multiplied by 100%, it becomes the rate of change in 𝑌

changes when 𝑋 increase by 1 unit.
LOG-LIN MODEL

16
 LIN-LOG MODELS

» Lin-log models follow this general form:

𝑌𝑖 = 𝛽0 + 𝛽1 ln 𝑋𝑖 + 𝑢𝑖

» Note that 𝛽1 is the change in Y responding to some percentage (or relative)

change in X
• If X increases by 100%, predicted Y increases by 𝛽1 units.
 EXPLAINING THE LIN-LOG MODEL
» The model:
𝑌 = 𝛽0 + 𝛽1 ln 𝑋 + 𝜀
1
» Slope (marginal effect): 𝛽1
𝑋
1
When 𝑋 increases by 1 unit, 𝑌 changes by 𝛽1 .
𝑋
The marginal effect in a log-linear model depends on the current value of 𝑋.

1
» Elasticity: 𝛽1
𝑌
1
When 𝑋 increase by 1%, 𝑌 changes by 𝛽1 percent.
𝑌
The elasticity in a log-linear model depends on the current value of 𝑌.
» The meaning of β1 : increase in 𝑌 when 𝑋 doubles.
LIN-LOG MODEL

19
A SUMMARY: FUNCTION FORMS
MODEL FORM MARGINAL EFFECT
𝜕𝑌 𝜕𝑌
𝜕𝑋 ELASTICITY 𝑌
𝜕𝑋
𝑋

𝑋
Linear 𝑌 = 𝛽0 + 𝛽1 𝑋 𝛽1 𝛽1
𝑌

𝑌
Log-linear 𝑙𝑛𝑌 = 𝛽0 + 𝛽1 𝑙𝑛𝑋 𝛽1 𝛽1
𝑋

Log-lin 𝑙𝑛𝑌 = 𝛽0 + 𝛽1 𝑋 𝛽1 𝑌 𝛽1 𝑋

1 1
Lin-log 𝑌 = 𝛽0 + 𝛽1 𝑙𝑛𝑋 𝛽1 𝛽1
𝑋 𝑌
20
QUADRATIC FUNCTIONAL FORM

21
 EXAMPLE DATA
A survey of 20,306 individuals in the U.S.
» male 1 = male; 2 = female
» age age (year)
» wage wage ($/hour)
» tenure # years working for current employer
» union 1 = union member, 0 otherwise
» edu years of schooling (years)
» race 1 = white; 2 = black; 3 = others
» married 1 = married or living together with a partner, 0 otherwise
Data file: qlr.xlsx
DATA DESCRIPTION
 MINCER’S WAGE FUNCTION…

» ... is a function of schooling years and experience

ln 𝑤𝑎𝑔𝑒 = 𝛽0 + 𝛽1 𝑒𝑑𝑢 + 𝛽2 𝑡𝑒𝑛𝑢𝑟𝑒 + 𝛾𝑋 + 𝜀

where X is a set of individual characteristics.

THE WAGE FUNCTION ESTIMATED
 THE SQUARED EXPLANATORY VARIABLE
» It is hypothesized that wage increases with age, but up to some point of age,
wage starts decreasing. How would you test this hypothesis?
 THE WAGE FUNCTION
WITH SQUARED REGRESSOR
QUALITATIVE REGRESSORS
Dummy variable as a regressor
Transforming categorical variables into dummies
Dummy regressors and structural change/difference

28
DUMMY VARIABLE AS A REGRESSOR
Coefficient of a dummy regressor
should be interpreted as the
difference between the two groups
of the dummy regressor.
 INTRODUCING CATEGORICAL VARIABLE

» Recall the variable race, which is a categorical variable:

• race = 1 if white, = 2 if black, 3 if others.

» How to introduce this variable into the wage function?

» We can’t introduce it directly to the regression function.
» Instead, we have to create a set of corresponding dummy variables

30
 TRANSFORMING CATEGORICAL VARIABLES TO DUMMY VARIABLES
race = 1 if white, = 2 if black, 3 if others
» We generate
• white: = 1 if race == 1, 0 otherwise
• black: = 1 if race == 2, 0 otherwise
» We introduce white and black to the regression
» others is then considered base category

31
THE WAGE FUNCTION WITH
CATEGORICAL VARIABLES
β of white/black indicates the
difference in ln(wage) between
white/black and the base
category (“others”).
 GENDER DIFFERENCE
IN RETURN TO EDUCATION

» If we want to test whether there is a difference in return to education between

male and female
MODEL WITH INTERACTION TERM
 RACE DIFFERENCE
IN RETURN TO EDUCATION

» Now test whether there is difference in return to education between racial

groups (white, black and others)
 RACIAL DIFFERENCE IN
RETURN TO EDUCATION

β of edu_white indicates the

difference in the coefficient of
edu between white and others.