EC325 Metrics Flashcards
Random experiment example
Tennessee Project STAR - Krueger
Measuring effect of class size on educational achievement
11600 students and teachers randomly assigned to small regular or regular with a full time teachers aide classes
Found that being in a smaller class was associated with a 5% increase in test scores
7 Problems when running experiments
- Randomization bias: people selecting to take part in the trial may have inherent difference to other people
- Supply side changes: in the trial the supply side may be more motivated than during actual implementation
- Attrition bias: attrition rates (leaving sample group) may be different between treatment and control groups
- Hawthorne effects: people behave differently if they’re part of an experiment
- Contamination bias: subjects in control group might nevertheless get treated
- Substitution bias: control group members may seek substitutes for treatment
- Externalities: within treatment group or between groups
D in D example
Card &; Krueger: Effect of minimum wage increase in New Jersey
Control group: Pennsylvania (no wage increase)
Treatment: New Jersey (increase from 4.25 to 5.05)
found no effect
D in D Assumption
Parallel trends!! can’t prove but can test through historic data
IV
Find a variable that increases the probability of ending up in the treatment group but which is uncorrelated with any other variable that influences the outcome
Assumptions:
Variable affects probability of treatment (first stage)
Does not affect outcome except by affecting probability of treatment (exclusion restriction)
AFFECTS PROBABILITY OF TREATMENT BUT HAS NO DIRECT EFFECT ON OUTCOME
IV Example
Angrist &; Grueger used quarter of birth as an IV for schooling.
They find that an extra year of education is associated with a 9% increase in weekly earnings
RD assumptions
Conditional on the running variable, the potential outcomes are continuous at the discontinuity, usually true when:
- No other relevant policies change at the discontinuity
- Running variable cannot be precisely manipulated (self selection not possible)
Example of RD
Lee uses a sharp RD to estimate probability that an incumbent wins election
Large increase in probability of winning if you are incumbent
UI Effects Experiment
Problem as UI schemes are usually national so empirical identification is difficult, however in US UI is a state programme with large variation so can use state reforms as NATURAL EXPERIMENTS with treatments (those affected by reform) and controls (those not affected)
- Johnston and Mas estimate that a one month reduction in UI leads to a 0.45 month reduction in UI claiming spells and 0.25 month reduction in unemployment duration
RAND HIE
The RAND Health Insurance Experiment was designed to address the issue of selection bias occurring due to health insurance not being randomly assigned.
It was a social experiment in the 1970s across 4 cities. Individuals were randomly assigned into different coinsurance groups ranging from 5% to 95% coverage with a $1000 cap on expenses.
Findings:
- Healthcare IS price sensitive (elasticity around 0.2)
- Those with more generous coverage didn’t see a health improvement
How does Medicaid affect health
Takeup: 70%
Crowd Out: 20-50%
Currie-Gruber used sophisticated diff-in-diff methods exploiting the variation in eligibility across 50 states to find that Medicaid for young children reduces child mortality and that medicaid expansions for pregnant women reduced the incidence of infant mortality and low birth weight
Empirical Evidence of Crowd-Out
Gruber & Hungerman - IV approach
Andreoni & Payne - IV approach
Both found that crowd out is less than 1
Every $1 donated by government to charity led to 0.56$ less private donations
Empirical evidence for Net of tax wages and high skilled migration
Kleven-Landaiz-Saez
Estimated effect of tax rate on mobility of top football players in Europe
Used variations in top marginal tax rate (good approximate of average wage for top footballers) in Europe as natural experiments and used a diff in diff approach
One example is ‘Beckham Law’ in Spain that reduced top marginal tax rate to a flat rate of 24% for workers moving to Spain after 1st Jan 2004, for up to 6 years.
Treatment: Spain
Control: Synthetic Spain
Results:
- The elasticity of location with respect to the net-of-tax rate is 0.4 on the whole sample (overall migration effect)
- Elasticity is much larger for foreign than for domestic players
- Location elasticities are very large at the top of the ability distribution (about 2), but negative at the bottom (ability sorting effect)
- Cross-effect between foreign and domestic players is negative (displacement effect)
Elissa & Liebmann EITC Paper
Estimating the impact of EITC expansion on labour force participation and hours worked
Diff in Diff approach
Control - Single women without children
Treatment - Single mothers (targets of the tax credit)
Find large extensive margin response but not much intensive margin response
Problem with single women as the control is that their participation is already high so little room for any upward trend, so they used single women without children and LOW EDUCATION
Nichols-Zeckhauser result
If differences in demand across rich and poor are only due to income effects (luxuries vs necessities) then in kind transfers cannot improve target efficiency
If conditional on income demand is higher for low ability individuals then in kind transfers can improve target efficiency
