Experimental Design Details
This survey experiment will be implemented among 9,000 online respondents in nine Latin American countries (Colombia, Peru, Brazil, Chile, Argentina, Mexico, Costa Rica, Guatemala, Panama), with 1,000 respondents targeted in each of the 9 countries. These countries span different ideological orientations, average levels of trust in government, and types of subsidy programs. The online survey will be implemented by LAPOP. The sample will be drawn from a standing online panel, which is representative of the online-population in these countries.
The experimental design randomly varied whether individual respondents were exposed to the placebo, T1, or T2 vignette. The vignettes were designed to vary the institutions responsible for implementing a transfer to compensate poor households for the higher prices that come with eliminating subsidies. Following the vignette, we measured an attitudinal outcome for respondents’ level of support for subsidy reform. .
We will analyze the average treatment effects of (1) introducing compensation for reform and (2) of varying the institution responsible for implementing the compensation using a specification as follows:
Y_it=β_0+β_1 〖T1,2〗_i+γX_(i,t-1)+δ_c+ϵ_ic (1)
Y_it=β_0+β_1 〖T1〗_i+β_2 〖T2〗_i+γX_(i,t-1)+δ_c+ϵ_ic (2)
where i indexes women, c indexes countries, and t indexes whether the measurement is pre- or post-treatment (t-1 is pre-treatment, t is post-treatment). Equation (1) pools treatments 1 and 2 to test the overall effect of introducing compensation for reform, regardless of implementation designation. Treatment effects are relative to the base group of assignment to the placebo group. Equation (2) separately tests the effects of T1 and T2. We would then test the difference between β_1 and β_2 in Equation (2) to test the effect of assigning implementation to local governments, churches, and NGOs rather than to national government. Both specifications include country fixed effects. X_(i,t-1) is a vector of control variables, including: educational attainment, income level, gender, age, frequency of public transportation use, and support for current president / head of state.
In addition to testing the effects in the pooled sample across all countries, we may test treatment effects in specific countries in order to support more specific policy recommendations and analysis relevant to countries considering reforms.