Acknowledgements
1
Initial conditions
Metrics is marketing
Know yourself
The best science comes from people who know themselves
So, what is your empirical prior?
My objectives include challenging your priors
The bottom line…
Healthy skepticism
Course objectives
Logistics
2
The big picture
On approach
The role of assumption
Two dimensions to validity
Internal validity: Should the study be believed?
External validity: Can the study’s results be reasonably generalized?
So, what to do when randomization is not possible?
3
Difference estimators
4
OLS in review
Population vs. sample
Linear regression
The estimator
Being the “best”
OLS
vs.
other lines/estimators
OLS formally
OLS properties and assumptions
What properties might we care about for an estimator?
So, what properties might we care about for an estimator?
Answer 1: Bias
Answer 2: Variance
The bias-variance tradeoff
OLS Properties
OLS Assumptions
Uncertainty and inference
Learning from our errors
Confidence intervals
Hypothesis testing
F
tests
5
Notes on OLS
Why the constant term?
What matters more, though…
For example
The problem with OLS?
Omitted-variable bias
OVB in one picture
6
Matching
Example 1: Exact matching
More generally…
Example 2
Types of matching
Exact matching
Inexact matching
Propensity score matching
Nearest-neighbour matching
Kernel matching
Still-other types of matching
Bandwidth selection
The basics
Cross-validation
Choosing a matching method
The common-support problem
Variable selection in matching
Balancing tests
Standard errors
The problem
Solutions
Example 3
Matching in Stata
Econometrics
Summary
Literature
7
Difference-in-differences
General setup
An example
Ashenfelter (1978)
DD with panel data
8
Simulation
Simulating DD in R
The Function
The grids that cover parameter space
The simulation itself
The presentation of results
Simulating DD in Stata
Seting up the panel
The program
The simulation
The presentation of results
Can we simulate a dip?
ECL plots
The triple difference
9
Instrumental variables
General discussion
IV in pictures
Example: Project Star
The just-identified case
Deriving the estimator
The Wald estimator
Another example (distance as IV)
Running IV models
IV in STATA
IV in R
Testing exogeneity
Method 1: Using first-stage regression residuals
Method 2: A
\(t\)
-test on the coefficient of interest
Durbin-Wu-Hausman test in STATA
Testing over-identifying restrictions
Simulating IV (in R)
OLS
2SLS
In summary
Correct SE estimates in 2SLS (in R)
Card (1993)
Twins and returns to schooling
References
Weak instruments
Heterogeneous treatment effects
The linear model
Local average treatment effects
Revisting the distance example
Heterogeneous effects: Multiple instruments
Where to get IVs
Literature
10
Regression discontinuity
General discussion
Sharp Regression Discontinuity (SRD)
Fuzzy Regression Discontinuity (FRD)
Other RD issues
RD Presentation
Specification testing:
RD Bandwith
“Guide to Practice”
Case 1: SRD designs
Case 2: FRD designs
RD in Stata
Method 1
Method 2
Method 3
RD in R
Simulated
Example
McCrary tests
Donut RD
Literature
11
Synthetic controls
Running synnthetic controls in R
12
Bounding exercises
13
Standard-error estimation
Simulations of Linear Regression
Setting Up Simulations
Data without Clusters
Data with Clusters
Cluster-Robust SE, Fixed Effects, or Random Effects Models
14
Manuscript review
Internal validity: Should the study be believed?
External validity: Can the study’s results be reasonably generalized?
Literature
Micro-Econometrics
Chapter 12
Bounding exercises
More to come…