Scribd
Upload
19 views

Lecture 6

The document discusses the Multiple Regression Model (MRM) and its advantages over the Single Regression Model (SRM), including improved prediction accuracy and the ability to account for omitted variables. It explains key concepts such as partial and marginal slopes, adjusted R-squared, and omitted variable bias, along with their implications in regression analysis. The lecture also touches on the importance of randomized control trials in estimating treatment effects.

Uploaded by

eythanam
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
19 views

Lecture 6

The document discusses the Multiple Regression Model (MRM) and its advantages over the Single Regression Model (SRM), including improved prediction accuracy and the ability to account for omitted variables. It explains key concepts such as partial and marginal slopes, adjusted R-squared, and omitted variable bias, along with their implications in regression analysis. The lecture also touches on the importance of randomized control trials in estimating treatment effects.

Uploaded by

eythanam
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 21

Global Poverty and

Impact Evaluation

Econ 174
Lecture 6
■ Multiple Regression Model (MRM)

Today’s Agenda ○

○
Interpret the coefficients of an MRM
Distinguish between a partial slope and
marginal slope
○ Understand omitted variable bias
○ Assess the overall fit of the regression
using adjusted R2
MULTIPLE ■ The MRM extends the SRM by
incorporating other explanatory
REGRESSION variables in its equation.
MRM is important because it allows
MODEL (MRM)
■
us to:
1. Make more accurate
predictions than the SRM.
2. Account for important omitted
variables.
3. Learn about what factors drive
our response variable.
SRM VS. MRM
Single Regression Model Multiple Regression Model

■ The unobserved errors in the model ■ The unobserved errors in the model
have mean 0 and have mean 0 and
○ are independent of one another. ○ are independent of one another.
○ have equal variance. ○ have equal variance.
○ are normally distributed around ○ are normally distributed around
the regression line. the regression line.
○ 𝐸 𝜖𝑥 =0 ○ 𝐸 𝜖 𝑥1 , … , 𝑥𝑘 = 0
Democracy and 𝑑𝑒𝑚𝑜𝑐𝑟𝑎𝑐𝑦 = 𝛽0 + 𝛽1 𝑠𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔 + 𝜖
Education
■ A key assumption about 𝜖 is that it
represents random variation.
■ 𝜖 does not depend on the other
independent variables (exogeneity
assumption)

■ 𝑐𝑜𝑣 𝑠𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔, 𝜖 = 0
SINGLE REGRESSION MODEL
෣
𝑑𝑒𝑚𝑜𝑐𝑟𝑎𝑐𝑦 = 0.156 + 0.038 ∗ 𝑠𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔
MULTIPLE
By adding GDP per capita to the
REGRESSION regression model, the population
MODEL regression line is,

𝑑𝑒𝑚𝑜𝑐𝑟𝑎𝑐𝑦
= 𝛽0 + 𝛽1 𝑠𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔 + 𝛽2 𝐺𝐷𝑃𝑝𝑐 + 𝜖
■ Three parameters ( , , ) to
estimate using OLS
MULTIPLE REGRESSION MODEL
෣
𝑑𝑒𝑚𝑜𝑐𝑟𝑎𝑐𝑦 = 0.242 + 0.023 ∗ 𝑠𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔 + 0.33 ∗ 𝐺𝐷𝑃𝑝𝑐
Marginal vs Partial Partial slope: Slope of an explanatory variable
in a multiple regression
Slopes ■ Statistically exclude the effects of other
explanatory variables
Marginal slope: Slope of an explanatory
variable in a simple regression
■ May combine the direct effect of the
explanatory variable with the effects from
other factors
Partial and marginal slopes are the same only
when the explanatory variables are uncorrelated.
• R2: A measure of goodness-
Adjusted R-squared of-fit
• Measures what proportion of
the variability of y is
explained by variation in the
prediction line
• R2 increases when including
any explanatory variable to
the regression model, even if
it is unrelated to the
response variable.
Adjusted R2:
Adjusted R-squared Adjusts for the sample size n and model
size k (where k is the number of
explanatory variables)
Is always smaller than R2

Measures how much variation in the


response variable is explained by the
prediction line
Omitted Variable • OVB occurs when a relevant
explanatory variable is not included in
Bias a regression model.

• It can cause the coefficient of one or


more explanatory variables to be
biased.

• MRM helps with OVB by including the


missing variable in the regression.
Omitted Variable • But sometimes, we can’t
include all relevant variables.
Bias
• We need to examine how the
missing variable might affect
results.

• OVB formula allows us to sign


the bias created by an omitted
variable.
OMITTED VARIABLE BIAS FORMULA
(+)
(+)

X and omitted X and omitted


variable are positively variable are
correlated negatively correlated
Omitted variable is
positively correlated Positive Bias
with Y
Omitted variable is
negatively correlated
with Y
OMITTED VARIABLE BIAS FORMULA
(–)
(+)

X and omitted X and omitted


variable are positively variable are
correlated negatively correlated
Omitted variable is
positively correlated Positive Bias Negative Bias
with Y
Omitted variable is
negatively correlated
with Y
OMITTED VARIABLE BIAS FORMULA
(+)
(–)

X and omitted X and omitted


variable are positively variable are
correlated negatively correlated
Omitted variable is
positively correlated Positive Bias Negative Bias
with Y
Omitted variable is
negatively correlated Negative Bias
with Y
OMITTED VARIABLE BIAS FORMULA
(–)
(–)

X and omitted X and omitted


variable are positively variable are
correlated negatively correlated
Omitted variable is
positively correlated Positive Bias Negative Bias
with Y
Omitted variable is
negatively correlated Negative Bias Positive Bias
with Y
OMITTED VARIABLE BIAS FORMULA
(+) (+)
𝑐𝑜𝑣(𝑆𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔, 𝐺𝐷𝑃𝑝𝑐)
𝑏𝑆𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔 = 𝛽𝑠𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔 + 𝛽𝐺𝐷𝑃𝑝𝑐
𝑣𝑎𝑟(𝑆𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔)

X and omitted X and omitted


variable are positively variable are
correlated negatively correlated
Omitted variable is
positively correlated Positive Bias Negative Bias
with Y
Omitted variable is
negatively correlated Negative Bias Positive Bias
with Y
OMITTED VARIABLE BIAS
■ Suppose we have the following
Randomized Control regression model:

Trials 𝑦𝑖 = 𝛽0 + 𝛽1 𝐷𝑖 + 𝛽2 𝑋𝑖 + 𝜖𝑖

■ 𝐷𝑖 is an indicator for being


(randomly) assigned to treatment.

■ Suppose we estimate instead

■ 𝑦𝑖 = 𝛾0 + 𝛾1 𝐷𝑖 + 𝜖𝑖

■ How do 𝛾1 and 𝛽1 compare?

■ Because of random assignment


𝐶𝑜𝑣 𝐷𝑖 , 𝜖𝑖 = 0
Next Class
• Regressions with indicator
variables
• Interaction Effects

You might also like