Please enable JavaScript.
Coggle requires JavaScript to display documents.
Quant II 1 & 2 Regression with Panel Data (Evidence Based Research…
Quant II 1 & 2
Regression with Panel Data
Control unobservable variables
Vary time, fixed location: automobile safety innovation
Fixed time, vary location: drinking culture
Balanced panel has all observation for each entity and each time period.
This method can be used for unbalanced data in certain regression software.
"Before & After"
By assuming that an OLS is fixed in time, using After - Before formula will remove the OLS from the equation.
Only applies when Time = 2. If there are more time data, use Fixed Effects Regression.
Fixed Effects Regression
Fixed time, vary location
Time Fixed Effects
Same over location, time-invariant variable
Entity Fixed Effects
Same over time, entity-invariant variable
Read 10.6 at Page 418
One variable must be fixed. Unfortunately panel data is often unavailable.
For varying time and entities, Instrumental Variables Regression is used.
Previous Prob with OLS:
Internal Validity
Reverse causality
Omitted Variable Bias (OVB)
Wrong functional form
External Validity
Generalisability
Goal:
To at least be able to tell between good and bad analysis
Result Chain
Input
Activities
Output
Outcome
Final Outcome
To produce a rigorous
Theories of Change
is basis for Hypothesis Testing
Evidence Based Research
Indonesia fertility
: improve resource allocation (educating women reduces fertility rate)
Progressa:
policy outcome evaluation and political sustainability (giving money to the poor ensures their education and health. program continued despite being previous government's idea)
Colombia malnourishment & cognitive:
improve policy design, assessing alternative
Kenya deworming, uniform, meal, school attendance:
evaluate cost-effectiveness of several methods
Mexico happy cement floor:
result chain
Types of Impact Assessment
Cause-&-Effect:
how intervention can change outcome
Normative:
comparing what's happening and what should be happening
Descriptive:
tells what's happening
To research or not?
Innovative
Replicable
Strategically relevant
Untested
Influential
Selecting KPIs
S
pecific
M
easurable
A
ttributable
R
ealistic
T
argeted
As growth increase, public wages as a share of GDP will decrease
Assumption of Regression Analysis
Sample is IID (Independent Identical Distribution)
Expected value of error = 0
No outlier, can check thru scatterplot
