Weighted 2SLS spillover effects

The task

Task statement

Use households.csv and census.csv. Keep all households with recipient=0. This includes ineligible households in both village treatment statuses. Fit weighted 2SLS with two endogenous columns dose*(1-eligible) and dose*eligible. Exogenous columns are 1-eligible, eligible, baseline*(1-eligible), and baseline*eligible. Instruments are these four columns plus instrument*(1-eligible) and instrument*eligible. Add no other intercept. Use weight as the regression weight. For each eligibility group, the spillover effect is its dose slope times the household-weighted mean dose among its retained observations. Report ineligible_effect and eligible_nonrecipient_effect. Average these two effects using nonrecipient_population from census.csv. Report population_effect. The counterfactual sets dose to zero for every retained household. Census totals determine only the final between-group weights. For every result, n_obs is the number of nonrecipient household rows used in estimation. The group effect equals the fitted dose slope times its weighted mean dose. Attribution to a zero-dose counterfactual is a stated synthetic convention.

Data

households.csv: household = Synthetic household identifier; eligible = Eligibility-group indicator, zero or one; treated = Own village treatment indicator; recipient = Indicator for actually receiving the transfer; instrument = Supplied excluded instrument for dose; dose = Supplied treatment exposure; baseline = Supplied baseline covariate; weight = Positive observation weight for estimation and within-group averages; outcome = Synthetic dependent variable in the stated regression census.csv: eligible = Eligibility-group indicator, zero or one; nonrecipient_population = Census count used only for final group aggregation

Files

input.csv, households.csv, census.csv

Required output

Scored estimates: ineligible_effect, eligible_nonrecipient_effect, population_effect.

Reference output

A run succeeds when every scored estimate lies within 1 percent or 0.005 of its reference value, whichever is larger. Sample sizes are not scored.