Econ 244, Lecture III: Sample Selection Models
Chris Walters
University of California, Berkeley
September 25, 2015
Introduction
Heckit
Equivalence
Extrapolation
LIV
Multiple Treatments
Introduction
This lecture discusses selection models
We will also disc
Regression Discontinuity (and Kink) Designs
The Regression Discontinuity Design (RDD) has become one of the most popular quasiexperimental
research designs in applied economics. This is because of the methods transparency and the lack of reliance
on para
Panel Data II
In this lecture we will review the material in Chamberlain (1984)s famous handbook chapter on panel data
and briey cover some dynamic models. We start with correlated random eects estimation of linear models
and then discuss dynamic panel mo
Hazard Models
This nal lecture provides a brief introduction to hazard models, which are statistical models for explaining spell lengths (e.g. unemployment duration or the length of an individuals life). Two statistical problems
are endemic to spell data.
Panel Data I
In this lecture we will review linear panel data models based upon variance component representations of
the DGP. We will then discuss a number of practical issues that frequently arise in estimating such models
and some common research desig
Regression Discontinuity II
Pat Kline
Topics
Special topics in RDD
Quantile RD
Multiple Discontinuities
RKD
Fuzzy / Sharp
Estimation
Inference
Beyond Means
Standard RDD can be used to identify
distribution of potential outcomes at cutoff.
Simple
Linear Regression Model
y = X

+
i ~ iid(0, 2) ! E() = 0, Var() = 2INxN
E(Xiki) = 0 for all k
X has full column rank K
^
^
ols = ( X X )1 X y , E ols =
^ ^
Var ols = 2 ( X X ) 1 ,
^
^2
=
^ ^
n K
=
RSS
n K
^
ols is BLUE
Statistical problems with
2
University of California, Berkeley
Department of Economics
Ken Chay
Fall Semester, 2005
ECON 244
APPLIED EXERCISE #3
Due 4 pm on November 21, In Ken Chays mailbox
This exercise examines the following research question: What is the impact of Chiles 900 Sch
Lecture: Selection on Observables
(Hand in P.S. #1)
Evaluation/Selection Problem:
Ex. Linear additive model
yi = + Ti + X i + i
Focus on binary (01) treatment, homogeneous treatment effects
1, if treated
Ti =
0 , otherwise
i = i
i has 2 potential outc
Estimation Principles
In this lecture we will review general estimation principles which extend far beyond OLS to general
nonlinear setups. The most fundamental results are for extremum estimators formed by minimizing or
maximizing an objective function.
Regression as Reduction
In order to arrive at a distinct formulation of statistical problems, it is necessary to dene
the task which the statistician sets himself: briey, and in its most concrete form, the object
of statistical methods is the reduction of
Structure, Design, and Causality
This lecture introduces basic concepts in econometric modeling and their relationship to modern notions
of causality. In the process, we lay the foundation for many of the subjects that will be covered in this class.
1
Pre
Econ 244, Lecture I: Regression, Selection on
Observables, and Quantiles
Chris Walters
University of California, Berkeley
September 18, 2015
Introduction
Regression
Selection on Observables
Distributions
Introduction
This class surveys methods in modern m
Duration Models
This final lecture provides a brief introduction to duration models (sometimes called hazard models),
which are statistical models for explaining spell lengths (e.g. unemployment duration or the length of an
individuals life). The early li
Econ 244, Lecture IV: Regression Discontinuity
Chris Walters
University of California, Berkeley
October 2, 2015
Introduction
Sharp/Fuzzy RD
Diagnostics
Estimation/Inference
Examples
Introduction
The regression discontinuity design (RD) is one of the most
Econ 244, Lecture II: Instrumental Variables
Chris Walters
University of California, Berkeley
September 18, 2015
Introduction
Constant Effects
Heterogeneous Effects
Introduction
This lecture covers instrumental variables (IV) methods
IV has been at the ce
Panel Data I
In this lecture we review linear panel data models based upon variance component representations of the
DGP. We then discuss a number of practical issues that frequently arise in estimating such models and some
common research designs predica
Panel Data II
In this lecture we will review the material in Chamberlain (1984)s famous handbook chapter on panel data
and briefly cover some dynamic models. We start with correlated random effects estimation of linear models
and then discuss dynamic pane
Structure, Design, and Causality
This lecture introduces basic concepts in econometric modeling and their relationship to modern notions
of causality. In the process, we lay the foundation for many of the subjects that will be covered in this class.
1
Pre
Estimation Principles
In this lecture we will review general estimation principles. It is worth reading this lecture along with
Newey and McFadden (1994)s famous Handbook of Econometrics chapter (from which I borrow liberally)
to deepen your understanding
Homework #1
Due in class 9/19/14
Please work in groups of 24
Answers must be typed
1. Identication I (OLS). Consider the linear regression model:
Yi = Xi + ui
where Yi and ui are scalar random variables and Xi is a K 1 random vector. The structures in th
Homework #2
Due in class 10/10/14
Answers Must be Typed
Work in Groups of 24
1) Consider the following Table from Fehr and Goette (2007) reporting descriptive statistics from an
individuallevel randomized experiment:
Focus on treatment period 1 where th
Homework #3
Due 11/21/14
1) (Reweighting) Go to David Autors webpage and download the cleaned
1979 and 1997 MORG les from:
http:/econwww.mit.edu/faculty/dautor/data/autkatkear08
a) Read through the cleaning programs associated with the MORG les.
Describe
Homework #4
Due in class 12/15/14
1) (CRE vs Fixed Eects)
Consider the model
Yit = i + Xit + it
where i is a xed person eect, Xit is a scalar regressor, and it is a strictly exogenous time varying error.
Assume the data are balanced, with T observations o