0% found this document useful (0 votes)
8 views

Lecture 3 - Functional Forms

The document discusses various functional forms and qualitative regressors in regression analysis, including linear, log-linear, lin-log, and quadratic forms. It utilizes data from Vietnam's provinces to illustrate these concepts, focusing on how different variables, including dummy and categorical variables, impact GDP and wages. Additionally, it examines the significance of interaction terms in assessing differences in returns to education across gender and racial groups.
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
8 views

Lecture 3 - Functional Forms

The document discusses various functional forms and qualitative regressors in regression analysis, including linear, log-linear, lin-log, and quadratic forms. It utilizes data from Vietnam's provinces to illustrate these concepts, focusing on how different variables, including dummy and categorical variables, impact GDP and wages. Additionally, it examines the significance of interaction terms in assessing differences in returns to education across gender and racial groups.
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 31

FUNCTIONAL

FORMS AND
QUALITATIVE
REGRESSORS
Nguyen Quang
[email protected]
OUTLINE
• Dummy variables
• Categorical variables
• Functional forms
• Linear
• Log-linear
• Log-lin
• Lin-log
• Quadratic functional form
• Interaction terms
DATA
Viet Nam Provincial data (58 provinces, 5 years, with some missing values)
• gdp: provincial GDP (bil. VND)
• labfo: number of laborers of provinces (1000 persons)
• rinvest: gross investment of provinces (bil. VND)
• year: 2007 – 2011
• city_ucg: 1 = city under the central government; 0 = province
• zone: economic zone (NMA – Northern mountainous area; RRD – Red River
Delta; NCC – North Central and Central Coast; CHL – Central Highlands; SE -
Southeast; MRD – Mekong River Delta)
File: fnf.xlsx
DATA IMPORT
• Consider a linear regression function
𝑌 = 𝛽! + 𝛽"𝑋 + 𝜀

• 𝛽" is the change in Y when X increases by 1 unit.


LINEAR AND • Sometimes the relationship is not linear.
SOME OTHER
FUNCTIONAL • Common functional form:
FORMS • Linear

• Log-linear
• Log-lin
• Lin-log
FUNCTIONAL FORMS
Linear Log-linear

𝑌 = 𝛽! + 𝛽" 𝑋 + 𝜀 ln 𝑌 = 𝛽! + 𝛽" ln 𝑋 + 𝜀

Lin-log Log-lin

𝑌 = 𝛽! + 𝛽" ln 𝑋 + 𝜀 ln 𝑌 = 𝛽! + 𝛽" 𝑋 + 𝜀
VIET NAM PROVINCIAL DATA DESCRIPTION

*nma, ncc, chl, se and mrd are dummies for the zone categorical variable
DATA DESCRIPTION
DATA DESCRIPTION
EXPLAINING THE LINEAR MODEL
• Consider a linear regression function
𝑌 = 𝛽! + 𝛽"𝑋 + 𝜀

• Slope (marginal effect): 𝛽"


• When 𝑋 increases by 1 unit, 𝑌 changes by 𝛽"

#
• Elasticity: 𝛽"
$

#
• When 𝑋 increase by 1%, 𝑌 changes by 𝛽" percent.
$

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


REGRSSION:
LINEAR
MODEL
LOG-LINEAR MODEL
• Example - The Cobb-Douglas Production Function:
𝑄% = 𝛽"𝐿% &! 𝐾% &"

can be transformed into a linear model by taking natural logs of both sides:
ln 𝑄% = ln 𝛽" + 𝛽' ln 𝐿% + 𝛽( ln 𝐾%

• The slope coefficients can be interpreted as elasticities.


EXPLAINING THE LOG-LINEAR MODEL

• Consider a regression with log-linear functional form:


ln 𝑌 = 𝛽! + 𝛽" ln 𝑋 + 𝜀

$ $
• Slope (marginal effect): 𝛽" à When 𝑋 increases by 1 unit, 𝑌 changes by 𝛽" .
# #

• The marginal effect in a log-linear model depends on the values of 𝑋 and 𝑌.

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

• Exact calculation: When 𝑋 increase by 1%, 𝑌 changes by 𝟏. 𝟎𝟏𝜷𝟏 − 𝟏 ×𝟏𝟎𝟎 percent.


LOG-LINEAR MODEL

• Approximation: 1% increase in labor force


leads to 0.505% increase in Real GDP
• Exact calculation: 1% increase in labor force
leads to 1.01!.#!# − 1 ×100 = 0.504%
increase in Real GDP

• What if 𝛽*$%&(()*+,) = 20 (which is a


substantially greater magnitude)?
• What if labor force is insignificant in this
model?
LOG-LINEAR MODEL

Adding the dummy for Cities Under the


Central Government (CUCG)
• Approximation: Compared to provinces, the
real GDP of CUCG is 26.78% higher.
• Exact calculation: Compared to provinces, the
real GDP of CUCG is 𝒆𝟎.𝟐𝟔𝟕𝟖 − 𝟏 ×𝟏𝟎𝟎 =
𝟑𝟎. 𝟕𝟏% higher. Note the difference from the
case of a continuous variable like 𝑙𝑎𝑏𝑓𝑜.
• What if city_ucg is insignificant in this model?
LOG-LINEAR MODEL

Adding dummies for economic regions


(categorical variable ‘zone’), including nma, ncc,
chl, se and mrd to the regression.
• β of each dummy indicates the difference in real
GDP between that region and the base category,
‘rrd’ (Red River Delta).
LOG-LIN MODEL

• Example – The growth model of real GDP with growth rate 𝑟:


𝑅𝐺𝐷𝑃+ = 𝑅𝐺𝐷𝑃!(1 + 𝑟)+

can be transformed into a linear regression by taking natural logs of both sides:
ln 𝑅𝐺𝐷𝑃+ = ln 𝑅𝐺𝐷𝑃! + 𝑡 ln(1 + 𝑟)

• Let 𝛽! = ln 𝑅𝐺𝐷𝑃! and 𝛽" = ln(1 + 𝑟), this can be rewritten as:
ln 𝑅𝐺𝐷𝑃+ = 𝛽! + 𝛽"𝑡
EXPLAINING THE LOG-LIN MODEL
• The model:
ln 𝑌 = 𝛽! + 𝛽"𝑋 + 𝜀
• Slope (marginal effect): 𝛽"𝑌 à When 𝑋 increases by 1 unit, 𝑌 changes by 𝛽"𝑌.
• The marginal effect in a log-linear model depends on the current value of 𝑌.
• Elasticity: 𝛽"𝑋 à When 𝑋 increase by 1%, 𝑌 changes by 𝛽"𝑋 percent.
• The elasticity in a log-linear model depends on the current value of 𝑋.
• The meaning of 𝜷𝟏 :
• Approximation: When X increase by 1 unit, Y changes by 𝜷𝟏 ∗ 𝟏𝟎𝟎 percent
• Exact calculation: When X increase by 1 unit, Y changes by (𝐞𝛃𝟏 − 𝟏) ∗ 𝟏𝟎𝟎 percent
LOG-LIN
MODEL
• Take 𝛽34)5 for an example:
• Approximation: Real GDP
increases by 8.17% per year.
• Exact calculation: Real GDP
increases by 𝒆𝟎.𝟎𝟖𝟏𝟕 − 𝟏 ∗
𝟏𝟎𝟎 = 𝟖. 𝟓𝟏% per year
• What about the case of dummies and
categorical variables?
THE LIN-LOG MODEL
• The model:
𝑌 = 𝛽! + 𝛽" ln 𝑋 + 𝜀
" "
• Slope (marginal effect): 𝛽" # à When 𝑋 increases by 1 unit, 𝑌 changes by 𝛽" #.
• The marginal effect in a log-linear model depends on the current value of 𝑋.
" "
• Elasticity: 𝛽" $ à When 𝑋 increase by 1%, 𝑌 changes by 𝛽" $ percent.
• The elasticity in a log-linear model depends on the current value of 𝑌.
• The meaning of 𝜷𝟏 :
• Approximation: If X increases by 1%, Y increases by 𝜷𝟏 /𝟏𝟎𝟎 units.

• Exact calculation: If X increases by 1%, Y increases by 𝜷𝟏 𝐥𝐧(𝟏. 𝟎𝟏) units.


LIN-LOG
MODEL
• Take 𝛽$%&(()*+,) for an example:
• Approximation: If labor force
increase 1%, real GDP
increases by 61.29 bil. VND.
• Exact calculation: If labor force
increase 1%, real GDP
increases by 𝟔𝟏𝟐𝟗×
𝐥 𝐧 𝟏. 𝟎𝟏 = 𝟔𝟎. 𝟗𝟗 bil. VND.
• What about the case of dummies
and categorical variables?
QUADRATIC
FUNCTIONAL FORM
THE SQUARED • It is hypothesized that wage increases with age, but up to
EXPLANATORY some point of age, wage starts decreasing.
VARIABLE • How would you test this hypothesis?
DATA
• A survey of 20,306 individuals in the US:
• wage: wage ($/hour)
• edu: years of schooling (years)
• age: age (years)
• tenure: years working for current employer
• union: 1 = union member, 0 otherwise
• male: 1 = male; 2 = female
• married: 1 = married/living with a partner, 0 otherwise
• race: 1 = white; 2 = black; 3 = others
• File: qlr.xlsx
WAGE DATA
DESCRIPTION
Note the use of I(age^2)
THE WAGE
FUNCTION
WITH SQUARED
REGRESSOR
• t-test of the squared variable I(age^2) helps
compare quadratic and linear functional forms
• H0: the coefficient of I(age^2) = 0. t=-6.357; p-
value<10%; reject H0. Conclusion: The quadratic
functional form represents the impact of age on
wage better than the linear functional form.
THE WAGE
FUNCTION
WITH SQUARED
REGRESSOR
• The quadratic relationship
𝑦 = 𝛽! + 𝛽" 𝑋 + 𝛽# 𝑋 #
takes on a U-shape or an inverse U-shape
depending on the sign of the coefficient of the
squared term (𝛽# ).
• The function reaches its maximum or minimum
$!
at 𝑋 = − . Whether this point lies within the
#$"
value range of X in the data determines the
actual shape of the relationship.
GENDER DIFFERENCE
IN RETURN TO EDUCATION
If we want to test whether there is a difference in return to education
between male and female

Using 𝒆𝒅𝒖 ∗ 𝒎𝒂𝒍𝒆 introduces the


following regressors into the
regression model: 𝒆𝒅𝒖, 𝒎𝒂𝒍𝒆, and
the interaction term 𝒆𝒅𝒖×𝒎𝒂𝒍𝒆.
MODEL WITH
INTERACTION TERM

• The coefficient of 𝑒𝑑𝑢: 𝑚𝑎𝑙𝑒


indicates the difference in the impact
of the number of schooling years
on wages between males and
females.
• t-test of 𝑒𝑑𝑢: 𝑚𝑎𝑙𝑒 determines
whether the difference is significant.
RACE DIFFERENCE • Now test whether there is difference in return to
IN RETURN TO education between racial groups (white, black and others)
EDUCATION
RACIAL
DIFFERENCE IN
RETURN TO
EDUCATION

• The coefficient of
𝑒𝑑𝑢: 𝑤ℎ𝑖𝑡𝑒 indicates the
difference in the impact of
the number of schooling
years on wages between
white people and those
from the ‘others’ group.
• t-test of 𝑒𝑑𝑢: 𝑤ℎ𝑖𝑡𝑒
determines whether the
difference is significant.

You might also like