Scribd
Upload
293 views

Applied Econometrics Notes

This document summarizes key concepts from an applied econometrics lecture. 1) The most important assumption of ordinary least squares (OLS) regression is that the independent variables are uncorrelated with the error term. Violations of this assumption, such as omitted variable bias or reversed causality, can lead to endogeneity. 2) Randomized experiments allow identification of causal effects because random assignment ensures the independent variables are uncorrelated with the error term, unlike observational data. 3) It is essential that treatment and control groups in an experiment are assigned randomly to ensure internal and external validity. Control variables may still be needed if randomization depends on covariates.

Uploaded by

Nga Nguyen
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
293 views

Applied Econometrics Notes

This document summarizes key concepts from an applied econometrics lecture. 1) The most important assumption of ordinary least squares (OLS) regression is that the independent variables are uncorrelated with the error term. Violations of this assumption, such as omitted variable bias or reversed causality, can lead to endogeneity. 2) Randomized experiments allow identification of causal effects because random assignment ensures the independent variables are uncorrelated with the error term, unlike observational data. 3) It is essential that treatment and control groups in an experiment are assigned randomly to ensure internal and external validity. Control variables may still be needed if randomization depends on covariates.

Uploaded by

Nga Nguyen
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 3

Applied Econometrics Lecture 1

- Key assumptions OLS



# Conditional mean of zero (Most important assumption):
X is independent of the error term. X is not influenced by other factors.
# Key concept 2.3: Know the covariance rules!

- If the sample becomes very large, the estimator of B1 becomes the true B1 plus
the error term. If conditional mean of the estimator is zero, then it follows that
the estimator is consistent and the estimator of B1 equals the true value of the
population.
# Violations of the OLS assumption endogeneity:
Omitted variable bias and reversed causality.
- Average treatment effect: The difference between all the Y1s minus the Y0s.
I.e. the average profit of a firm that implements the strategy minus the average
profit of a firm that does not implemenet the strategy.
# Counterfactual problem: We cannot calculate the average treatment
effect because the counterfactual outcomes are not observed. OLS can be
used torun a regression analysis, however the estimator will not be equal
to the true causal effect.

- The error term illustrate whether you are below or above average. If this is
related to X, then it will mismeasure the causal effect. The key assumption of OLS
is violated.
- Identification using experimental data;
# because of random assignment X is independent of the error term,
unlike observational data, which you always have to defend this
assumption

# Spillovers: Control and treatment group influence eachother (no
internal validity)

# General equilibrium effect: (No external validity)
- Most essential part of experiments is the random assignment. Thus no control
variables are needed, however randomization can depend on covariates.
# If the distribution of the two groups differ, i.e. major vs non major, then
you need to include a control variable to control for this. Conditional
mean does not hold, but conditional independence does.
- Control variables not needed:
# unless assignment is random conditional on W


CMI, X is not related to the error term. We cannot drop Wi in the formula,
because it is not randomly assigned, has no a causal interpretation. It still
can be related to omitted factors

# Precision: Lower standard errors by including other explanatory
variables

# Heterogeneous effects: Interaction effect. Do not forget to include the
W1 variable itself.
- Bad control can introduce endogeneity problem. Look at slide 38. If Y2 is
introduced, through Y2 X is related to the error term.

You might also like