Econometrı́a I

Multiple regression analysis: Further issues

UPB

Semestre II - 2022

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 1 / 14

 Functional Form

Functional forms

Change of scale
∆y , change in the scale of the coefficients and standard errors. No
change in significance or interpretation
∆xk , change in the scale of that coefficients and standard errors. No
change in significance or interpretation
Functional form (Nonlinear)
Logartithm, Log (y ), Log (xk )
Quadratic, xk2
Interactions, x̸=k ∗ xk

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 2 / 14

 Functional Form

Log models

Why?
Invariant to the scale of the variables since measuring percent changes
Direct estimate of the elasticity
Reduce heteroskedasticity or skewness
More narrow distribution, limiting the effect of outliers
Interpretation
ln(y ) = β0 + β1 ln(x) + u
β1 , the elasticity of y with respect to x
ln(y ) = β0 + β1 x + u
β1 , percentage change in y given a 1 unit change in x
y = β0 + β1 ln(x) + u
β1 , change in y for a 100 percent change in x

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 3 / 14

 Functional Form

Quadratic Models

y = β0 + β1 x + β2 x 2 + u

βˆ2 > 0, convex

βˆ2 < 0, concave

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 4 / 14

 Functional Form

Quadratic Models

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 5 / 14

 Functional Form

Significance tests: quadratic models

H0 : x has constant effect on y

H0 : x has no effect on y
H1 : x effect on y is not
H1 : x has effect on y
constant

Meaning:
Meaning:
H0 : βx = 0, βx 2 = 0
H0 : β x = 0
H1 : H0 , not true
H1 : H0 , not true

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 6 / 14

 Functional Form

Models with interaction terms

y = β0 + β1 x1 + β2 x2 + β3 x1 x2 + u

δˆ1 = βˆ1 + βˆ3 x2

δˆ1 = βˆ2 + βˆ3 x1
x = x̄
y = α0 + δ1 x1 + δ2 x2 + δ3 (x1 − x¯1 )(x2 − x¯2 ) + u

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 7 / 14

 Functional Form

Adjusted R-Squared

R 2 will always increase as more variables are added to the model

Adjusted R 2 takes into account the number of variables in a model
and may decrease
SSR/(n − k − 1) σˆ2
R¯2 ≡ 1 − =1−
SST /(n − 1) SST /(n − 1)

Useful for comparing fit of two models with same y

Not to fixate too much on R 2 or R¯2
If an economic theory predicts a variable belongs, generally leave it in

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 8 / 14

 Qualitative information

Qualitative factor

Categorical
Any categorical can be changed to a dummy variable
Dichotomic or binary variable
A dummy variable is a variable that takes on the value 1 or 0
yi = β0 + β1 di + ui yi = β0 + β1 di + β2 xi + ui

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 9 / 14

 Qualitative information

Dummy variables

yi = β0 + β1 di + β2 xi + ui

di = 1 + di = 0 = 1
β1 = E (yi |di = 1, xi ) − E (yi |di = 0, xi )
di = 1, Interpretation group
di = 0, Base group

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 10 / 14

 Qualitative information

Dummy variables for multiple categories

yi = β0 + β1 d1i + β2 d2i + β3 xi + ui

Dummy variable trap

Not include all the groups of the same category
Avoid trap: C (categories) - 1 in the regression
H0 : β1 = β2 , not possible to test directly

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 11 / 14

 Qualitative information

Interactions involving dummy variables

yi = β0 + β1 d1i + β2 d2i + β3 d1i ∗ d2i + β4 xi + ui

Interacting dummy variables is like subdividing the group

Different intercept
Show yi if changes in d1i d2i

yi = β0 + β1 d1i + β2 xi + β3 d1i ∗ xi + ui

Interacting dummy variables is like pondering one group

Different slope
Show yi if changes in d1i

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 12 / 14

 Qualitative information

Testing for dummies: Chow test

Estimate the model with all the interactions (unrestricted) and
without (restricted). Then form an F statistic
y = β0 + β1 x1 + ... + βk xk + (δ0 d + δ1 d ∗ x1 + ... + δk d ∗ xk ) + u

H0 : δ0 , δ1 , ..., δk
SSRr − SSRur /(k + 1)
F ≡
SSRur /(n − 2(k + 1))

With no constant, δ0
SSRr − SSRur /k
F ≡
SSRur /(n − 2(k + 1))

Restricted (Pooled): SSRp

Unrestricted: SSR1 + SSR2
SSRp − (SSR1 + SSR2 )/k + 1
F ≡
(SSR1 + SSR2 )/(n − 2(k + 1))
Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 13 / 14
 Qualitative information

A Binary Dependent Variable (EXTRA)

yi = (0, 1)
E (y = 1|x) = β0 + β1 x1 + ... + βk xk

Probability of success
Linear probability model (LPM)
Coefficients: Increase or decrease in the probability of success
A probability cannot be linearly related to the independent variables
for all their possible values
Potential problem that can be outside [0,1]
Violates assumption of homoskedasticity

Multiple regression analysis: Further issues (UPB) Econometrı́a I Semestre II - 2022 14 / 14