Econometrı́a I
Análisis de regresión multiple
UPB
Semestre II - 2022
Análisis de regresión multiple (UPB) Econometrı́a I Semestre II - 2022 1 / 12
Multiple regression
Multiple regression
Model with two independent variables
yi
xi1
xi2
β0
β1
β2
ui
Model with k independent variables
Análisis de regresión multiple (UPB) Econometrı́a I Semestre II - 2022 2 / 12
Multiple regression
Interpretation
ŷ = βˆ0 + βˆk xk
Variation
Partial effect, ceteris paribus
Partial-out
1 Regress x , over intercept and all other variables and obtein the
j
residual
ˆ
xˆj = γ0 + γˆ−j x−j
rˆj = xj − xˆj
2 Regress y P
on the residual with no intercept
n
rˆij yi
βˆ1 = Pi=1
n 2
i=1 rˆij
Análisis de regresión multiple (UPB) Econometrı́a I Semestre II - 2022 3 / 12
Multiple regression
Partialling out
The residual is uncorrelated with the other explanatory variables.
The residual explicitly controls for the other explanatory variables by
removing any correlation they have
The slope coefficient represents the isolated effect
Simple estimate vs Multiple estimate
yˆi = βˆ0 + βˆ1 xˆ1
yˆi = γˆ0 + γˆ1 xˆ1 + γˆ2 xˆ2
Análisis de regresión multiple (UPB) Econometrı́a I Semestre II - 2022 4 / 12
Multiple regression
Goodness-of-Fit
(yi − ȳ )2
P
Total sum of squares (SST),
Explained sum of squares (SSE), (ŷi − ȳ )2
P
P 2
Residual sum of squares (SSR), ûi
SST = SSE + SSR
R 2 can never decrease when another independent variable is added
It is not a good way to compare models
Análisis de regresión multiple (UPB) Econometrı́a I Semestre II - 2022 5 / 12
Multiple regression
Unbiasedness assumptions
Required assumptions
Linear parameters
Random sampling
No perfect collinearity
Zero conditional mean
Análisis de regresión multiple (UPB) Econometrı́a I Semestre II - 2022 6 / 12
Multiple regression
Selection of variables
Too many variables
Include not related variables
No effect on our parameters
OLS remains unbiased
Too few variables
Exclude related variables
Effects over our parameters
OLS will usually be biased
Análisis de regresión multiple (UPB) Econometrı́a I Semestre II - 2022 7 / 12
Multiple regression
Omitted variable bias
True model Ommited model
y = β0 + β1 x1 + β2 x2 + u y = β0 + β1 x1 + v
β̃1,short = β̂1,long + β̂2,long δ̂1
Análisis de regresión multiple (UPB) Econometrı́a I Semestre II - 2022 8 / 12
Multiple regression
Omitted variable bias
Bias = 0 :
β2 = 0
Corr (x1 , x2 ) = 0
Bias ̸= 0 :
Análisis de regresión multiple (UPB) Econometrı́a I Semestre II - 2022 9 / 12
Multiple regression
Variance of OLS estimators
Homoskedasticity
The error u has the same variance given any value of the explanatory
variables
Var (u|x1 , ..., xk ) = σ 2
Var (y |x1 , ..., xk ) = σ 2
Gauss-Markov assumptions (5)
Assumptions for unbiasedness
Linear parameters
Random sampling
No perfect collinearity
Zero conditional mean
Homoskedasticity assumption
Análisis de regresión multiple (UPB) Econometrı́a I Semestre II - 2022 10 / 12
Multiple regression
Variance of OLS estimator
σ2
Var (β̂j ) =
SSTj (1 − Rj2 )
Components
σ2
(xij − x̄j )2
P
SSTj =
Rj2 , regressing xj on all other x’s
Standard deviation q
ˆ (β̂j |x
se(β̂j ) = Var
Análisis de regresión multiple (UPB) Econometrı́a I Semestre II - 2022 11 / 12
Multiple regression
Efficiency of OLS: The Gauss-Markov Theorem
Under Gauss-Markov the OLS estimator β̂j for βj is the best linear
unbiased estimator
BLUE, Best Linear Unbiased Estimator
Best, having the smallest variance
Linear, can be expressed as a linear function of the data on the
dependent variable
n
X
β̃j = wij yi
i=1
Unbiased, assumptions
Análisis de regresión multiple (UPB) Econometrı́a I Semestre II - 2022 12 / 12
