Qualitative

Variables &
Functional
Form
Dr. Henry
Kankwamba

Functional
form
Logarithms
Polynomials Qualitative Variables & Functional Form
Qualitative
Variables

Dr. Henry Kankwamba

Econometrics AAE 316

October 10, 2022

 Qualitative
Variables &
Functional
Form
Dr. Henry
Kankwamba

Functional
form
Logarithms
1 Functional form
Polynomials

Qualitative
Logarithms
Variables
Polynomials

2 Qualitative Variables
 Qualitative
Variables &
Functional
Form
References
Dr. Henry
Kankwamba

Functional
form
Logarithms
Polynomials

Qualitative
Variables

• Wooldridge chapters 6 & 7

• Stock and Watson chapter 8
 Qualitative
Variables &
Functional
Form
Dr. Henry
Kankwamba

Functional
form
Logarithms
Polynomials

Qualitative
Section 1
Variables

Functional form
 Qualitative
Variables &
Functional
Form
Functional form
Dr. Henry
Kankwamba

Functional
form
Logarithms
Polynomials • We can use functions of variables in regression models
Qualitative
Variables
• Example: wages, years of education, & experience
(a) (a) (a)
a Wagei = β0 + β1 Educi + β2 Experi + εi
(b) (b) (b)
b log Wagei = β0 + β1 Educi + β2 Experi + εi
(c) (c) (c) (c)
c log Wagei = β0 + β1 Educi + β2 Experi + β3 (Experi )2 εi
• All three models can be estimated by OLS
• What is the interpretation of β (a) vs β (b) vs β (c) ?
• How to choose which model to estimate?
 Qualitative
Variables &
Functional
Form
Logarithms
Dr. Henry
Kankwamba

Functional
form
Logarithms • We often estimate regressions using logarithms of
Polynomials

Qualitative
some variables instead of the level
Variables • Example: wages and years of education
• Level: Wagei = β0lev + β1lev Educi + εi
• Logarithm: log(Wagei ) = β0log + β1log Educi + εilog
• Interpretation of slope
• Each additional year of education is associated with an
increase in wages of β lev dollars
• Each additional year of education is associated with an
increase in log wages of β log
 Qualitative
Variables &
Functional
Form
Change in logarithm
Dr. Henry
Kankwamba
approximates percent change
Functional
form
Logarithms • If log w changes by ∆, then change the percent change
in w is
Polynomials

Qualitative
Variables
wnew − wold e (log w )+∆ − e log w
100 =100
wold e log w
e log w
=100 log w (e ∆ − 1)
e

=100 ∆ + ∆2 /2 + ∆3 /3! + ∆4 /4! + · · · Ta
≈100∆

• So 100β1log is the percentage change in wages

associated with an additional year of education
 Qualitative
Variables &
Functional
Form
Change in logarithm
Dr. Henry
Kankwamba
approximates percent change
Functional
form
Logarithms
Polynomials
• In,
Qualitative
Variables log:log log:log log:log
log(Wagei ) = β0 + β1 log(Educi ) + εi

• β1log:log is the change in log wage associated with a

change in log years of education
• A 1% change in Educ increases log(Educ) by 0.01, so
log:log
log(Wage) increases by 0.01β1 , or Wage increases
log:log
by β1 %
• β1log:log is the elasticity of wages with respect to
education
 Qualitative
Variables &
Functional
Form
Logs vs levels
Dr. Henry
Kankwamba

Functional
form
Logarithms
Polynomials

Qualitative
• When to use logarithms instead of levels?
Variables
• No universal rules
• Logarithms useful when:
• Percent changes are desirable
• Variable in levels is positive and right skewed e.g.
income, population, wealth
• Logarithm cannot be used with variables that are ≤ 0
 Qualitative
Variables &
Functional
Form
Polynomials
Dr. Henry
Kankwamba
• Example: wages, years of education, & experience
Functional (b) (b) (b)
form b log Wagei = β0 + β1 Educi + β2 Experi + εi
(c) (c) (c) (c)
Logarithms
Polynomials
c log Wagei = β0 + β1 Educi + β2 Experi + β3 (Experi )2 εi
Qualitative • Interpretation:
Variables
• β2(b) is the increase in log Wage associated with an
additional year of experience
• β2(c) is the increase in log Wage associated with an
additional year of experience holding Exper 2 fixed –
does not make sense
• β2(c) + 2β3(c) Exper is the average derivative of log Wage
with respect to experience
• When to use polynomials?
• Benefit: allows richer non-linear relationship
• Costs: more difficult to report, more parameters to
estimate
 Qualitative
Variables &
Functional
Form
Dr. Henry
Kankwamba

Functional
form
Logarithms
Polynomials
Section 2
Qualitative
Variables

Qualitative Variables
 Qualitative
Variables &
Functional
Form
Qualitative Variables
Dr. Henry
Kankwamba • Categorical variable is one that has one or more
categories, but there is no natural ordering to the
Functional
form categories
Logarithms
Polynomials
• Examples: Gender, race, marital status, geographic
Qualitative location.
Variables
• The following two variables are equivalent:


1 if observation i corresponds to a woman,
Genderi =
2 if observation i corresponds to a man.


1 if observation i corresponds to a man,
Genderi =
2 if observation i corresponds to a woman.

• Qualitative variables cannot be simply included in

regression, because their units are arbitrary
 Qualitative
Variables &
Functional
Form
Dummy variables
Dr. Henry
Kankwamba

Functional
form
Logarithms
Polynomials • A dummy variable is a binary zero-one variable which
Qualitative takes on the value one if some condition is satisfied
Variables
and zero if that condition fails
• Female
 i =
1 if observation i corresponds to a woman,
0 if observation i corresponds to a man.
Malei =
• 
1 if observation i corresponds to a man,
0 if observation i corresponds to a woman.
• Note that Femalei + Malei = 1 for all observations i
 Qualitative
Variables &
Functional
Form
A single dummy independent
Dr. Henry
Kankwamba
variable
• Consider the following regression:
Functional
form
Logarithms
Wagei = β0 +δ0 Femalei +β1 Educi +β3 Experi +β4 Tenurei +εi ,
Polynomials

Qualitative assume E[εi |Female, Educ, Exper , Tenure] = 0.

Variables • If observation i corresponds to a woman, Femalei = 1,
and

E[Wagei |Femalei = 1, Educi , Experi , Tenurei ] =

= β0 + δ0 + β1 Educi + β3 Experi + β4 Tenurei .

• If observation i corresponds to a man, Femalei = 0, and

E[Wagei |Femalei = 0, Educi , Experi , Tenurei ] =

= β0 + β1 Educi + β3 Experi + β4 Tenurei .

• Thus,

δ0 =E[Wagei |Femalei = 1, Educi , Experi , Tenurei ]−

− E[Wagei |Femalei = 0, Educi , Experi , Tenurei ]
 Qualitative
Variables &
Functional
Form
An intercept shift
Dr. Henry
Kankwamba
• The model:
Functional
form
Logarithms Wagei = β0 +δ0 Femalei +β1 Educi +β3 Experi +β4 Tenurei +εi
Polynomials

Qualitative
Variables
• For men (Femalei = 0):, we can write the model as

WageiM = β0 + β1 Educi + β3 Experi + β4 Tenurei + εi .

• For women (Femalei = 1):, we can write the model as

WageiF = (β0 + δ0 ) + β1 Educi + β3 Experi + β4 Tenurei + εi .

• In this case, men are called the reference (or base)

group
• δ0 measures the difference relative to the reference
group.
 Qualitative
Variables &
Functional
Form
An intercept shift
Dr. Henry
Kankwamba

Functional
form
Logarithms
Polynomials

Qualitative
Variables
 Qualitative
Variables &
Functional
Form
Example
Dr. Henry
Kankwamba

Functional
form
• Estimated equation:
Logarithms
Polynomials
\ i = − 1.57 − 1.81 Femalei + 0.572 Educi +
Wage
Qualitative
Variables (0.72) (0.26) (0.049)
+ 0.025 Experi + 0.141 Tenurei
(0.012) (0.021)

• The dependent variable is the wage per hour

• δ̂0 = −1.81 implies that a women earns $1.81 less per
hour than a man with the same level of education,
experience, and tenure. (These are 1976 wages.)
• The difference is also statistically significant.
 Qualitative
Variables &
Functional
Form
Logarithmic dependent variable
Dr. Henry • The model:
Kankwamba

Functional log (Wage) = β0 +δ0 Female+β1 Educ+β3 Exper +β4 Tenure+ε.

form
Logarithms
Polynomials
• In this case,
Qualitative    
Variables δ0 = log Wage F − log Wage M
 
Wage F
= log
Wage M

!
Wage M + Wage F − Wage M
= log
Wage M
 
Wage F − Wage M
= log 1 +
Wage M
Wage − Wage M
F
≈ .
Wage M
• When the dependent variable is in the log form, δ0 has
a percentage interpretation
 Qualitative
Variables &
Functional
Form
Example
Dr. Henry
Kankwamba

Functional • Estimated equation:

form
Logarithms

(Wagei ) =
log\ 0.417 − 0.297 Femalei + 0.080 Edu
Polynomials

Qualitative
Variables (0.099) (0.036) (0.007)
+ 0.029 Experi − 0.00058 Experi2 +
(0.005) (0.00010)
+ 0.032 Tenurei − 0.00059 Tenurei2 .
(0.007) (0.00023)

• δ̂0 = −0.297 implies that a woman in 1976 earned

29.7% less than a man with the same level of education,
experience and tenure.
 Qualitative
Variables &
Functional
Form
Changing the reference group
Dr. Henry • Instead of
Kankwamba

Functional log (Wagei ) = β0 +δ0 Femalei +β1 Educi +β3 Experi +β4 Tenurei +εi
form
Logarithms
Polynomials
consider:
Qualitative
Variables log (Wagei ) = θ0 +γ0 Malei +θ1 Educi +θ3 Experi +θ4 Tenurei +εi .

• Since Malei = 1 − Femalei ,

log (Wagei ) = θ0 + γ0 Malei + θ1 Educi + θ3 Experi + θ4 Tenurei + εi

= θ0 + γ0 (1 − Femalei ) + θ1 Educi + θ3 Experi + θ4 Tenur
= (θ0 + γ0 ) − γ0 Femalei + θ1 Educi + θ3 Experi + θ4 Tenu

• We conclude that δ0 = −γ0 , β0 = θ0 − δ0 , β1 = θ1 , and

etc.

log (Wagei ) = (β0 + δ0 )−δ0 Malei +β1 Educi +β3 Experi +β4 Tenurei +εi

• Changing the reference group has no effect on the

conclusions
 Qualitative
Variables &
Functional
Form
The dummy variable trap
Dr. Henry
Kankwamba

Functional
form
• Consider the equation:
Logarithms

log (Wagei ) =β0 + δ0 Femalei + γ0 Malei +

Polynomials

Qualitative
Variables
+ β1 Educi + β3 Experi + β4 Tenurei + εi

• The intercept is like a regressor that takes the value

one for all observations
• Since Femalei + Malei − 1 = 0 for all observations i, we
have perfect multicollinearity, and such an equation
cannot be estimated.
• Cannot include an intercept and dummies for all
groups!
 Qualitative
Variables &
Functional
Form
The dummy variable trap
Dr. Henry
Kankwamba
• One of the dummies has to be omitted and the
Functional
form corresponding group becomes the reference group:
Logarithms
Polynomials
• Men are the reference group:
Qualitative
Variables log (Wagei ) = β0 +δ0 Femalei +β1 Educi +β3 Experi +β4 Tenurei +

• Women are the reference group:

log (Wagei ) = θ0 +γ0 Malei +β1 Educi +β3 Experi +β4 Tenurei +εi

• Alternatively, one can include both dummies without

the intercept:

log (Wagei ) = π0 Femalei +π1 Malei +β1 Educi +β3 Experi +β4 Te

but this is less common

 Qualitative
Variables &
Functional
Form
A slope shift and interactions
Dr. Henry • Can allow the returns to education to be different for
Kankwamba
men and women:
Functional
form
Logarithms
log (Wagei ) =β0 + δ0 Femalei + β1 Educi + δ1 (Femalei · Educi ) +
Polynomials
+ β3 Experi + β4 Tenurei + εi .
Qualitative
Variables
• The variable (Femalei · Educi ) is called an interaction
• The equation for men (Femalei = 0):
 
log WageiM = β0 +β1 Educi +β3 Experi +β4 Tenurei +εi .

• The equation for women (Femalei = 1):

 
log WageiF = (β0 + δ0 ) + (β1 + δ1 ) Educi +
+ β3 Experi + β4 Tenurei + εi .

• δ1 is the difference in return to education between

women and men (the reference group) after controlling
for experience and tenure
 Qualitative
Variables &
Functional
Form
A slope shift
Dr. Henry
Kankwamba

Functional
form
Logarithms
Polynomials

Qualitative
Variables
 Qualitative
Variables &
Functional
Form
Example
Dr. Henry
Kankwamba
• Estimated equation:
Functional
form
Logarithms (Wagei ) =
log\ 0.389 − 0.227 Femalei +
Polynomials

Qualitative
(0.119) (0.168)
Variables
+ 0.082 Educi − 0.0056 Femalei · Educ
(0.008) (0.0131)
+ 0.029 Experi − 0.00058 Experi2 +
(0.005) (0.00011)
+ 0.032 Tenurei − 0.00059 Tenurei2 .
(0.007) (0.00024)

• δ̂1 = −0.0056 suggests that the return to education for

women is 0.56% less than for men, however it is not
statistically significant
 Qualitative
Variables &
Functional
Form
Multiple categories
Dr. Henry
Kankwamba
• In the previous examples, Educ was a quantitative
Functional
form
variable: years of education
Logarithms
Polynomials
• Suppose now that instead the education variable is
Qualitative ordinal :
Variables


 1 if high-school dropout,


 2 if high-school graduate,
Education = 3 if some college,



 4 if college graduate,

5 if advanced degree.

• Only the order is important, and there is no meaning

to the particular values
• Adding such a variable to the regression will give a
difficult to interpret result
 Qualitative
Variables &
Functional
Form
Multiple categories

Dr. Henry  1 if high-school dropout,
Kankwamba 


 2 if high-school graduate,
Educationi = 3 if some college,
Functional 


 4 if college graduate,
form 
5 if advanced degree.
Logarithms
Polynomials
• Define 5 new dummy variables:
Qualitative  
Variables 1 if high-school dropout, 1 if high-school graduate,
E1,i = E2,i =
0 otherwise. 0 otherwise.
 
1 if some college, 1 if college graduate,
E3,i = E4,i =
0 otherwise. 0 otherwise.

1 if advanced degree,
E5,i =
0 otherwise.

• To avoid multicollinearity, one of the dummies has to

be omitted:

Wagei = β0 +δ0 Femalei +δ2 E2,i +δ3 E3,i +δ4 E4,i +δ5 E5,i +Other Variables

• Group 1 (high-school dropout) becomes the reference

group.
• δ2 measures the wage difference between high-school
graduates and high-school dropouts
• δ3 measures the wage difference between individuals
with some college education and high-school dropouts
• δ3 − δ2 is the wage difference between some college
and high-school graduates