Quasi-Experimental Methods
E N D
Presentation Transcript
Quasi-Experimental Methods Florence Kondylis (World Bank)
Objective • Find a plausible counterfactual • Reality check • Every method is associated with an assumption • The stronger the assumption the more we need to worry about the causal effect • Question your assumptions
Program to evaluate Fertilizer vouchers Program (2007-08) • Main Objective • Increase maize production • Intervention: vouchers distribution • Target group: • Maize producers • Farmers owning >1 Ha, <3 Ha land • Indicator: Yield (Maize)
I. Before-after identification strategy Counterfactual: Yield before program started • EFFECT = After minus Before Counterfactual assumption: There is no other factor than the vouchers affecting yield from 2007 to 2008 years
Questioning the counterfactual assumption Question: what else might have happened in 2007-2008 to affect maize yield ?
Examine assumption with prior data Assumption of no change over time not so great ! >> There are external factors (rainfall, pests…)
II. Non-participant identification strategy Counterfactual: Rate of pregnancy among non-participants Counterfactual assumption: Without vouchers, participants would as productive as non-participants in a given year
Questioning the counterfactual assumption Question: how might participants differ from non-participants?
Test assumption with pre-program data REJECT counterfactual hypothesis of same productivity
III. Difference-in-Difference identification strategy Counterfactual: • Non-participant maize yield, purging pre-program differences between participants/nonparticipants • “Before vouchers” maize yield, purging before-after change for nonparticipants (external factors) • 1 and 2 are equivalent
Effect = 3.47 – 11.13 = - 7.66 Participants 66.37 – 62.90 = 3.47 57.50 - 46.37 = 11.13 Non-participants
Effect = 8.87 – 16.53 = - 7.66 Before 66.37 – 57.50 = 8.87 62.90 – 46.37 = 16.53 After
Counterfactual assumption: Without intervention participants and nonparticipants’ pregnancy rates follow same trends
74.0 16.5
74.0 -7.6
Questioning the assumption • Why might participants’ trends differ from that of nonparticipants?
Examine assumption with pre-program data counterfactual hypothesis of same trends doesn’t look so believable
IV. Matching with Difference-in-Difference identification strategy Counterfactual: Comparison group is constructed by pairing each program participant with a “similar” nonparticipant using larger dataset – creating a control group from similar (in observable ways) non-participants
Counterfactual assumption: Unobserved characteristics do not affect outcomes of interest Unobserved = things we cannot measure (e.g. ability) or things we left out of the dataset Question: how might participants differ from matched nonparticipants?
73.36 Effect = - 7.01 66.37 Matched nonparticipant Participant
Can only test assumptionwith experimental data • Studies that compare both methods (because they have experimental data) find that: • unobservables often matter! • direction of bias is unpredictable! Apply with care – think very hard about unobservables
Summary • Randomization requires minimal assumptions needed and procures intuitive estimates (sample means !) • Non-experimental requires assumptions that must be carefully assessed • More data-intensive
Example: Irrigation for rice producers + Enhanced Market Access • Impact of interest measured by: • Input use & repayment of irrigation fee • Rice yield • (Cash) income from rice • Non-rice cash income (spillovers to other value chains) • Data: 500 farmers in project area / 500 random sample farmers • Before & after treatment • Can’t randomize irrigation so what is the counterfactual?
Plausible counterfactuals • Random sample difference in difference • Are farmers outside the scheme on the same trajectory ? • Farmers in the vicinity of the scheme but not included in scheme • Selection of project area needs to be carefully documented (elevation…) • Proximity implies “just-outside farmers” might also benefit from enhanced market linkages • What do we want to measure? • Propensity score matching • Unobservables determining on-farm productivity ?