AADAPT Workshop Latin AmericaBrasilia, November 16-20, 2009
Making the Most out of Discontinuities
Laura Chioda
Introduction
Setting: we would like to evaluate policy interventions in an observational setting, i.e. when the analyst cannot manipulate the selection process.
In general: Individuals, households, villages, or other entities, are either exposed or not exposed to a “treatment” or “policy regime” and the two groups are not comparable because of selection.
When randomization is not feasible, how can we exploit implementation features of the program to “measure” its impact?
Answer: Quasi-experiments (see Florence’s presentation) and now Regression
Discontinuity Design.
Regression Discontinuity Designs
RRD is closer cousin of randomized experiments than other competitors
Major element in the toolkit for empirical research Now very fashionable but it goes back to the
early sixties went into hibernation for two decades experienced a renaissance with the new
millennium
RDD Example
Policy: US drinking age – if less than 21, alcohol consumption is illegal
Outcomes: alcohol consumption and mortality rate
Observation: The policy treats people aged 20 and 11 months and 29 days and 21 year olds differently. However, do we think that these individuals are inherently different?
Are 20 years & 11 months and 29 days olds less wise, less likely to go to parties than 21 year olds? Less obedient?
People born “few days apart” are treated differently, because of the arbitrary age cut off established by the law. However, we hardly think that few days or a month apart could really make a difference in terms of behaviors and attitudes towards alcohol
RDD Example (2)
Idea: use this policy rule to assign people to treatment and control groups:
treatment group: those who are 20 years and 11 months old control group: individuals who just turned 21 – Why is this a sensible assignment rule?It is as if people were assigned to treatment and control at random.
RDD Example (3)
RDD Example (4)
RDD Logic
General idea: assignment to the treatment depends, either completely or partly, on a continuous “score”, ranking (age in the previous case): potential beneficiaries are ordered by looking at the
score there is a cut-off point for “eligibility” – clearly
defined criterion determined ex-ante cut-off determines the assignment to the treatment
or no-treatment groups
These de facto assignments often arise from administrative decisions, where the incentives to participate are partly limited because of resource constraints, and transparent rules rather than discretion are used for the allocation of incentives
Example (2): vouchers Government offers vouchers for
fertilizer for small farmers. Eligibility rule based on plot size:
If plot less than 2 km2 then farmer receives vouchers
If plot bigger than 2 km2 then no voucher Size of plot not easily manipulable
over night, easy to measure and enforce (with admin data on size of plots)
Everyone below the eligibility cut-off receives vouchers.
Example: fuzzy designNow suppose that, for unknown
reasons, not all the eligibles farmers receive the voucher. Why? limited knowledge of the program (didn’t
know the program was happening) Voluntary participation (farmers who
take up are different from those who don’t along several dimensions)
The percentage of participants changes discontinuously at cut-off, from zero to less than 100%
0.2
5.5
.75
1
tre
atm
en
t pro
bab
ility
assignment variable
Sharp Design for Voucher receipt
0.2
5.5
.75
1
assignment variable
Fuzzy Design for Voucher receipt
Probability of Participation under Alternative Designs
100%
0%
75%
0%
Sharp and Fuzzy Discontinuities
Ideal setting: Sharp discontinuity the discontinuity precisely determines
treatment status ▪ e.g. ONLY people 21 and older drink alcohol!▪ Only small plot receive vouchers
Fuzzy discontinuitythe percentage of participants changes discontinuously at cut-off, but not from zero to 100%
▪ e.g. rules determine eligibility but amongst the small farmers there is only partial compliance / take-up
▪ Some people younger than 21 end up consuming alcohol and some older than 21 don’t consume at all
Internal Validity
General idea: as a result of the arbitrary cut off associated to a given policy, individuals to the immediate left and right of the cut-off are similar. Therefore, differences in alcohol consumption and
mortality can be thought of as determined by the policy.
Assumption (nothing else is happening): in the absence of the policy, we would not observe a discontinuity in the outcomes around the cut off.
We are assuming that there is nothing else going on around the same cut off that impacts our outcome of interest: 21 year olds can start drinking however the moment
they turn 21 they have to enroll in a “drinking responsibly” type seminar
Vouchers: there is another policy that gives equipment to farmers with plots larger than 2 km2.
Outcome Profile Before and After the Intervention
outc
om
e
assignment variable
Baseline
assignment variable
Follow-up
Outcome Profile Before and After the Intervention
outc
om
e
assignment variable
Baseline
assignment variable
Follow-up
different shape
External Validity
How general are the results? Counterfactual: individuals “marginally
excluded from benefits” (less than 21, plots less than 2km2)
Causal conclusions are limited to individuals, households, villages at the cut-off The effect estimated is for individuals
“marginally eligible for benefits” extrapolation beyond this point needs additional,
often unwarranted, assumptions (or multiple cut-offs)
Fuzzy designs exacerbate the problem
The “nuts and bolts” of implementing RDDs
A major advantage of the RDD over competitors lies in its transparency, as it can be illustrated using graphical methods
Requires many observations around cut-off (alternatively, one could down-weight observations away from the cut-off)
Why? Because only near the cut-off can we assume that people find themselves by chance to the left and to the right of the cut-off. Think about farmer who owns 1 km2 plot vs
farmer who owns 50 km2 plot or compare a 16 vs a 25 years old.
Graphical Analysiso
utc
om
e
assignment variable
Moving the goalpost
Natural-experiments are “naturally” occurring instances which approximate the properties of an experiment
RDDs share the same properties as an experiment locally at the cut-off
Thus “real-world” discontinuities are a gold mine for those fishing for natural experiments
Wrap Up
Modern econometrics views RDDs as a powerful tool to identify causal effects Pros: as good as experiments (around
the cut off) Cons: the estimated program effects are
representative only of households/villages near the cut off, which may not reflect entire population of interest.
Wrap Up
Can be used to design a prospective evaluation when randomization is not feasible The design applies to all means tested
programs Multiple cut-offs to enhance external
validityCan be used to evaluate ex-post
interventions using discontinuities as “natural experiments”.
Thank You
22